New creator
You have a phone or a laptop and one recording that sounds worse than you hoped it would.
AI Tools · Experimental
AI Tools for Creative Work: Where to Start, at Any Budget
This page starts from the job you need done, not the brand, and lists tools at four budgets for each of 17 jobs in audio and music, film and video, photo and image. It is for first-time creators and working professionals who want to know where to start. By the end you will have a shortlist, a free option to try first, and a clear view of what each tool fails at. Experimental: the field changes every few months. Reviewed 2026-10-08.
How this page works
We have not tested any of these tools. Every entry is checked against public sources, and the status line on each job says so. 39 of 48 tools have been checked against the maker's own page. The rest are marked "Not yet". No placement is paid for, and no link carries an affiliate code.
Budgets follow four bands. Free: No cost. Includes open-source models and free tiers. Free tiers often forbid commercial use or add watermarks. Starter: Under $25 a month, or under $100 once. Pro: $25 to $100 a month, or $100 to $500 once. Studio: Above those bands, hardware, or quoted per contract. Prices are bands, not quotes, because vendors change them. Check the maker's page before you buy.
The jobs
-
Clean up dialogue
- Good at
- Steady background noise such as air conditioning, hum and room tone under speech.
- Fails on
- Overlapping voices, music under speech, and heavy reverb, where it leaves artefacts on consonants.
- A person still has to
- Listen on headphones and speakers, decide how much noise to leave so the edit still sounds natural.
- Free
- Adobe Enhance Speech, Audacity OpenVINO AI effects
- Starter
- Descript Studio Sound
- Pro
- Antinode Clear (formerly Supertone Clear), Waves Clarity Vx Pro, Accentize dxRevive Pro, iZotope RX
- Studio
- CEDAR DNS (contract product)
Not tested by us. Checked against public sources. Read the full guide
-
Remove noise on a live call or stream
- Good at
- Keyboard clicks, fans and room noise during a call.
- Fails on
- Music, TV in the room and a second speaker close to the mic.
- A person still has to
- Check the result on a recording of the call, since you cannot hear what the other end hears.
- Free
- NVIDIA Broadcast
- Starter
- Krisp
- Pro
- Waves Clarity Vx Pro
- Studio
- CEDAR DNS (contract product)
Not tested by us. Checked against public sources. Read the full guide
-
Separate stems
- Good at
- Pulling vocals, drums and bass out of a finished stereo mix.
- Fails on
- Dense mixes and reverb tails, where bleed from one stem appears in another.
- A person still has to
- Judge the bleed by ear and check you hold the rights to the track before releasing any stem.
- Free
- Audacity OpenVINO AI effects, Ultimate Vocal Remover
- Starter
- Moises, LALAL.AI
- Pro
- iZotope RX, Steinberg SpectraLayers
- Studio
- AudioShake (contract product)
Not tested by us. Checked against public sources. Read the full guide
-
Transcribe and caption
- Good at
- Clear single-speaker speech in common languages.
- Fails on
- Crosstalk, strong accents, names and technical terms, which need a human pass.
- A person still has to
- Proofread names and numbers, and check caption timing against picture.
- Free
- Audacity OpenVINO AI effects, OpenAI Whisper
- Starter
- Descript Studio Sound
- Pro
- Avid Pro Tools Speech-to-Text
- Studio
- AI-Media LEXI (contract product)
Not tested by us. Checked against public sources. Read the full guide
-
Make a synthetic voice or narration
- Good at
- Neutral narration and short lines in a voice you have the right to use.
- Fails on
- Emotional delivery in drama, and long takes where pacing drifts.
- A person still has to
- Get written consent for any real voice and label the result. See the ethics and labelling pages.
- Starter
- ElevenLabs
- Studio
- Respeecher (contract product)
Not tested by us. Checked against public sources. Read the full guide
-
Upscale and restore footage
- Good at
- Low-resolution or compressed footage with clean subjects and steady motion.
- Fails on
- Faces and fine text, where the tool invents detail the original never had.
- A person still has to
- Compare against the source frame by frame and say on screen the footage was enhanced.
- Free
- Upscayl
- Starter
- TensorPix
- Pro
- Topaz Video AI
- Studio
- Telestream Vantage Super Resolution (Announced, not yet released, contract product)
Not tested by us. Checked against public sources. Read the full guide
-
Fill, extend or remove in a photo
- Good at
- Removing small distractions and extending plain backgrounds.
- Fails on
- Repeating patterns, hands and anything the viewer knows well, such as a famous building.
- A person still has to
- Decide whether the edit is acceptable for the use. News and documentary work forbids it.
- Free
- Krita AI Diffusion
- Pro
- Adobe Photoshop Generative Fill
Not tested by us. Checked against public sources. Read the full guide
-
Generate video
- Good at
- Short clips of establishing shots, abstract motion and stylised scenes.
- Fails on
- Consistent characters across shots, readable text, and anything needing physical accuracy.
- A person still has to
- Check each clip frame by frame, check the licence terms, and label the result.
- Free
- Wan 2.2 in ComfyUI
- Starter
- Kling
- Pro
- Google Veo, Runway
- Studio
- Moonvalley Marey (contract product)
Not tested by us. Checked against public sources. Read the full guide
-
Split a finished mix into dialogue, music and effects
- Good at
- Pulling dialogue away from music and effects in a finished film, television or radio mix.
- Fails on
- Dense scenes where speech, music and effects share the same frequencies, which leave traces in the wrong stem.
- A person still has to
- Listen to each stem against the original, check you may reuse the music, and confirm what the delivery contract accepts.
- Pro
- Steinberg SpectraLayers
- Studio
- AudioShake (contract product)
Not tested by us. Checked against public sources. Read the full guide
-
Upmix stereo to surround or immersive
- Good at
- Spreading a stereo mix across surround channels when no stems survive.
- Fails on
- Mixes whose stereo width carries meaning, where an upmix changes the balance the mixer intended.
- A person still has to
- Check the downmix against the original and the result on calibrated speakers, and confirm the delivery spec allows an upmix.
- Pro
- NUGEN Halo Upmix
Not tested by us. Checked against public sources. Read the full guide
-
Speed up the edit with AI search, transcripts and auto cuts
- Good at
- Finding footage by description, turning speech into searchable text, and proposing a first multicam cut.
- Fails on
- Judgement about story and pacing, and any generated frame that has to match a real shot exactly.
- A person still has to
- Make every creative cut, check generated frames against the real footage, and confirm what the delivery allows.
- Pro
- Avid Media Composer, Adobe Premiere Pro, DaVinci Resolve Studio
Not tested by us. Checked against public sources. Read the full guide
-
Train or apply AI inside a VFX shot
- Good at
- Repeating a fix learned from a few hand-done frames across a shot, and relighting footage after the fact.
- Fails on
- Shots that differ from the training frames, and fine edge detail where the network fills in what it guesses.
- A person still has to
- Do the reference frames by hand, check every frame of the result, and keep the untouched plate.
- Free
- Autodesk Flow Studio
- Pro
- Beeble
- Studio
- Foundry Nuke CopyCat (contract product)
Not tested by us. Checked against public sources. Read the full guide
-
Restore and remaster damaged film and video
- Good at
- Detecting dust and scratches, and recreating a missing or badly damaged frame from its neighbours.
- Fails on
- Long damaged stretches, where a generated frame invents detail the camera never recorded.
- A person still has to
- Decide how much invention an archive accepts, keep the original scan, and document every generated frame.
- Pro
- Topaz Video AI
- Studio
- MTI Film DRS Nova (contract product), Phoenix restoration software (filmworkz.com) (contract product)
Not tested by us. Checked against public sources. Read the full guide
-
Caption live broadcasts with speech recognition
- Good at
- Producing a first caption stream from clean speech with no delay for a human to type, at a volume no crew could staff.
- Fails on
- Names, overlapping talkers, crowd noise and sound the engine has not met, where a wrong word can change a sentence.
- A person still has to
- Load names and terms before the show, watch the output live, correct or take over, and decide when quality is too low to air.
- Studio
- AI-Media LEXI (contract product), Verbit (VITAC Captivate) (contract product)
Not tested by us. Checked against public sources. Read the full guide
-
Subtitle or dub live commentary into other languages
- Good at
- Turning live speech into text in one language, and offering a second language stream that no interpreter booth was booked for.
- Fails on
- Idiom, names and fast commentary. Each language adds a second place for an error, after the speech step.
- A person still has to
- Choose which events and languages are worth the risk, test with real commentary, and keep a way to cut to the original.
- Studio
- AI-Media LEXI (contract product), CAMB.AI (contract product)
Not tested by us. Checked against public sources. Read the full guide
-
Check files before delivery and search a media archive
- Good at
- Flagging likely faults across more files than anyone could watch, and letting staff search a catalogue in plain language.
- Fails on
- Findings that look sure and are wrong, and shots the index never described. The makers' pages give no error rates.
- A person still has to
- Set the pass and fail thresholds, review each flag before rejecting a file, and confirm a found clip is the right one.
- Studio
- Interra Systems BATON (contract product), Telestream Qualify (contract product), Moments Lab (contract product)
Not tested by us. Checked against public sources. Read the full guide
-
Fill radio links with AI voice tracks and schedule by plain request
- Good at
- Preparing spoken links, station IDs and local items ahead of airtime, and taking scheduling requests typed in everyday words.
- Fails on
- Local knowledge, live events and tone. Whether anyone reviews generated copy before it airs is not stated on the pages we read.
- A person still has to
- Write or approve the copy, decide when a synthetic voice is acceptable, tell listeners, and keep a person ready for breaking news.
- Studio
- ENCO aiTrack (contract product), RCS RCSAI in SelectorCloud (Announced, not yet released, contract product), Futuri CoHostAI (contract product)
Not tested by us. Checked against public sources. Read the full guide
Routes by who you are
Each route lists the jobs to look at first, in order.
Podcaster or radio producer
Your chores repeat every episode, so the free local options pay back fastest.
Musician
Stems are the main use. Check who owns the track before you release any stem.
Indie filmmaker
A small budget and a long task list. Start with the jobs that have a free tool.
Post professional
You need to know what changed this year and which claims have a source.
Broadcast or radio engineer
Most products at your level are quoted by contract and are not tested here.
Professional strands
Film, television and broadcast work has different tests: real-time processing, contracts and delivery specifications. These strands cover it, tool by tool.
- AI in Professional Audio Post-Production. Dialogue editors, re-recording mixers, localisation and broadcast engineers.
- AI in Cinema and Television Production. Editors, VFX artists, restoration colourists and post supervisors.
- AI in Broadcast: Live Captions, Live Translation and Quality Control. Broadcast engineers, captioning coordinators, sports producers, media managers and archive staff.
- AI in Radio: Voice Tracking and Scheduling. Programme directors, radio engineers, producers and station managers.
Every tool, with tier and hardware
Several free tools need a graphics card. The hardware column says what the maker states, and "Not checked" where it does not. Local means your files stay on your computer. Cloud means you upload them, which matters for unreleased work and anything under a confidentiality agreement.
| Tool | Jobs | Tier | Where it runs | Hardware | Checked against maker |
|---|---|---|---|---|---|
| Adobe Enhance Speech | Clean up dialogue | Free | Cloud, upload required | Not checked | Not yet |
| Audacity OpenVINO AI effects Page lists noise suppression, Whisper.cpp transcription and translation, and music separation (vocal and instrumental, or vocals, drums, bass, other). Runs 100% locally per the page. The page states no licence for the effects. macOS and Linux need a compile with no macOS instructions yet. | Clean up dialogue, Separate stems, Transcribe and caption | Free | Runs on your computer | Windows download only. No hardware figures stated. | 2026-10-08 |
| Descript Studio Sound Page says Studio Sound removes noise and regenerates the voice, and does not say where it runs. It is not on the Free plan. It is listed on the Hobbyist plan. | Clean up dialogue, Transcribe and caption | Starter | Not checked | Not checked | 2026-10-08 |
| Antinode Clear (formerly Supertone Clear) Now made by Antinode Audio. Three controls: voice, ambience and voice reverb. Real-time processing is listed. Where processing happens is not stated. | Clean up dialogue | Pro | Not checked | Windows 10 or later (64-bit) and macOS 10.13 or later, with Apple Silicon support. AU, VST3, VST and AAX. | 2026-10-08 |
| Waves Clarity Vx Pro Page says real-time voice noise reduction powered by Waves Neural Networks. Platforms and formats were not in the part of the page we read. | Clean up dialogue, Remove noise on a live call or stream | Pro | Not checked | Not checked | 2026-10-08 |
| Accentize dxRevive Pro Page says all processing happens locally without sending data to the cloud. It restores noisy, clipped, reverberant and band-limited speech using neural networks. | Clean up dialogue | Pro | Runs on your computer | Mac (Apple Silicon recommended) and Windows. VST3, AU and AAX. No special hardware needed, per the page. | 2026-10-08 |
| iZotope RX Page lists RX 12 with Dialogue Isolate using neural nets in offline and real-time modes. MusicTech (2 July 2026) reports iZotope was acquired by Boris FX, licences stay active and development continues. No exact sale date was given. Platforms not stated on the product page. | Clean up dialogue, Separate stems | Pro | Runs on your computer | Not checked | 2026-10-08 |
| CEDAR DNS (contract product) Page announces DNS 8S for live sound and broadcast where audio over IP is not required, and lists DNS One as a plug-in. It does not say whether DNS products use machine learning. That claim comes from a trade source on DNS 4. | Clean up dialogue, Remove noise on a live call or stream | Studio | Runs on your computer | Not checked | 2026-10-08 |
| NVIDIA Broadcast The download page does not say the app is free. The free tier placement rests on secondary reviews. | Remove noise on a live call or stream | Free | Runs on your computer | GeForce RTX 2060, Quadro RTX 3000, TITAN RTX, or higher | 2026-10-08 |
| Krisp Page describes noise, echo and cross-talk removal for calls on desktop and a Chrome extension. It does not say where processing happens. The mobile app has no noise cancellation yet. A free sign-up exists and the page lists no free limits. | Remove noise on a live call or stream | Starter | Not checked | Not checked | 2026-10-08 |
| Ultimate Vocal Remover | Separate stems | Free | Runs on your computer | GPU conversions need at least an NVIDIA GTX 1060 6 GB, 8 GB recommended. AMD support is limited. 64-bit Windows 10 or later, or macOS Big Sur or later. | 2026-10-08 |
| Moises Page lists up to 27 stem types, a browser web app, desktop and mobile apps, and a free plan with a limited number of uploads. Where processing happens is not stated. | Separate stems | Starter | Not checked | Not checked | 2026-10-08 |
| LALAL.AI Page lists vocal, drums, bass, guitar, piano, synth, strings and wind stems and a VST plug-in that runs locally in a DAW. The free plan is described as for testing, with no download of the full result. | Separate stems | Starter | Not checked | Not checked | 2026-10-08 |
| Steinberg SpectraLayers Steinberg's SpectraLayers 12 help page documents Unmix Soundtrack, which unmixes the selected layer into dialogue, effects and music, with Balanced, High and Best quality settings that trade time for quality. The page says it suits audio from trailers, movies, television and radio shows. It states no limitations and does not say which edition includes it. | Separate stems, Split a finished mix into dialogue, music and effects | Pro | Runs on your computer | Not checked | 2026-10-08 |
| AudioShake (contract product) Its dialogue, music and effects page lists dialogue, music and effects stems and says clean dialogue improves transcription and captioning accuracy. It says dubbing services use the technology, and does not mention M&E tracks or recovering lost stems. Access is a browser tool, a REST API and SDKs. Customer logos and accuracy figures are the vendor's own claims and were not checked. | Separate stems, Split a finished mix into dialogue, music and effects | Studio | Cloud, upload required | Not checked | 2026-10-08 |
| OpenAI Whisper | Transcribe and caption | Free | Runs on your computer | GPU memory by model: tiny and base about 1 GB, small about 2 GB, medium about 5 GB, turbo about 6 GB, large about 10 GB. The turbo model does not translate. | 2026-10-08 |
| Avid Pro Tools Speech-to-Text Per ProSoundWeb (30 April 2026), the 2026.4 release enhances Speech-to-Text: select clips by word, sentence or speaker, show transcription lanes per track, carry transcription data through AudioSuite, Commit, Consolidate and .ptxm export, and choose CPU or GPU for analysis. Available to Pro Tools Studio and Ultimate. Languages not stated. Avid's own page was unreachable. | Transcribe and caption | Pro | Not checked | Not checked | Not yet |
| NUGEN Halo Upmix Page lists upmix to LCR, 3.1, 4.0, 4.1, 5.0, 5.1, 6.0, 6.1, 7.0 and 7.1, switchable "neural network powered dialog isolation" for the centre channel, an exact mode for downmix compatibility, and an optional 3D extension for Atmos bed (7.1.2), Auro 3D (7.1.4) and Ambisonic output. The page does not say where processing runs. | Upmix stereo to surround or immersive | Pro | Runs on your computer | AAX, VST3, AU and AudioSuite, 64-bit. The optional 3D extension supports Pro Tools (AAX) and Nuendo (VST3) on Windows and macOS. | 2026-10-08 |
| Avid Media Composer Per Newsshooter, Media Composer 2026.8 adds Ready to Edit (agentic content preparation), an embedded Google Gemini panel, PhraseFindAI Multicam Auto Cut (selects angles by speaker and cuts at speech boundaries, keeping the edit editable), and Avid Content Core search. Available from 1 September. Avid's own page was unreachable. | Speed up the edit with AI search, transcripts and auto cuts | Pro | Not checked | Not checked | Not yet |
| Adobe Premiere Pro Page lists Generative Extend using the Adobe Firefly Video Model, Media Intelligence search in plain language, Speech to Text in over 18 languages with speaker distinction, and auto captions translatable into 27 languages. Generative features have limited complimentary usage with no amounts given. The page gives no version number and does not say whether the features run in the cloud or locally. | Speed up the edit with AI search, transcripts and auto cuts | Pro | Not checked | Not checked | 2026-10-08 |
| DaVinci Resolve Studio Page names DaVinci Resolve 21. It says Studio includes the free version plus the DaVinci AI Neural Engine, and lists Magic Mask and text based editing under Studio. IntelliSearch is named as a new AI tool and Voice Isolation and Music Remixer as Fairlight AI effects, with no edition stated for either. IntelliScript and transcription are not named. | Speed up the edit with AI search, transcripts and auto cuts | Pro | Runs on your computer | Mac, Windows and Linux. | 2026-10-08 |
| Foundry Nuke CopyCat (contract product) A Foundry article says CopyCat, new in Nuke 13.0, lets an artist train a neural network on the shot at hand from a small set of before and after images, then run the saved network through the Inference node on the rest of the sequence. It names deblur, upscale and beauty fixes. It does not say where training runs. | Train or apply AI inside a VFX shot | Studio | Not checked | Not checked | 2026-10-08 |
| Beeble Page lists relighting ("3D Relight") and SwitchX, which it describes as mask-guided relighting and replacement frame by frame. It lists BeebleCLOUD in a web app and Beeble Studio marked local, and says in its Enterprise section "We never train on your data" with a full commercial license. It does not mention a Nuke plug-in. | Train or apply AI inside a VFX shot | Pro | Not checked | Not checked | 2026-10-08 |
| Autodesk Flow Studio Autodesk's own page returned 403. A search summary of its product pages, which we did not open, says Flow Studio is cloud-based, formerly Wonder Studio, and turns footage into editable CG with markerless motion capture, camera tracking, masks and clean plates, with Maya, Blender, Unreal and USD export. The free tier placement is unconfirmed. | Train or apply AI inside a VFX shot | Free | Cloud, upload required | Not checked | Not yet |
| MTI Film DRS Nova (contract product) Page lists manual cleanup, automated dust and defect repair, dewarping, and MTai FrameGen, which it says uses generative AI to recreate missing frames and, in version 7.0, replace highly damaged frames with an interpolation model trained by MTI. FrameGen is bundled with annual support. Sold through resellers. Not tested by us. | Restore and remaster damaged film and video | Studio | Runs on your computer | Windows 10 or 11, 128 GB RAM minimum and 256 GB recommended, a workstation-class CPU and a 16 GB graphics card minimum. | 2026-10-08 |
| Phoenix restoration software (filmworkz.com) (contract product) The old Digital Vision address redirected to this page when we opened it. The page calls Phoenix professional film and video restoration and grading software for hands-on work, with automated repair from DVO Tools. It makes no AI or machine learning statement and gives no hardware requirements. Not tested by us. | Restore and remaster damaged film and video | Studio | Runs on your computer | Not checked | 2026-10-08 |
| AI-Media LEXI (contract product) Per Sports Video Group, 15 April 2026: LEXI Text and LEXI Voice Encoders are the first new encoder hardware in more than a decade, with 4K over 12G-SDI, support for SDI, IP and cloud workflows, LEXI Live Sync for configurable caption and translation timing, and a Hardware-as-a-Subscription model. The Voice Encoder adds AI sound separation and noise removal for voice translation. The article names no languages, speech engine, price or ship date. AI-Media's own page was unreachable. | Transcribe and caption, Caption live broadcasts with speech recognition, Subtitle or dub live commentary into other languages | Studio | Cloud, upload required | Not checked | Not yet |
| Verbit (VITAC Captivate) (contract product) Verbit's home page offers real-time captions "powered by Captivate" for meetings, depositions, broadcasts and classrooms and describes Captivate as customizable ASR. It does not say live captioners are human, gives no accuracy figure and does not say that VITAC is now Verbit. A 2022 Sports Video Group article on the launch says clients can add a human co-pilot to control the live ASR session and that specialists prepare the engine before each session. Both are the maker's own claims. | Caption live broadcasts with speech recognition | Studio | Cloud, upload required | Not checked | 2026-10-08 |
| CAMB.AI (contract product) The EBU release of 6 March 2026 says Eurovision Sport and CAMB.AI provided real-time speech-to-text subtitles for all six Paralympic Winter Games sports, without altering editorial content. It names no languages and makes no accuracy or human-review claim. CAMB.AI's page lists "Live Dubbing" and "150+ languages" with no detail, and does not mention voice cloning or latency. | Subtitle or dub live commentary into other languages | Studio | Cloud, upload required | Not checked | 2026-10-08 |
| Interra Systems BATON (contract product) Page lists format, quality, timecode, loudness and compliance checks, neural-network quality checks, audio language identification, artifact detection and PSE detection. It runs in the cloud, on premises or hybrid, with REST and XML-RPC APIs and many file and caption formats. The page gives no error rates. | Check files before delivery and search a media archive | Studio | Cloud or on your own servers | Not checked | 2026-10-08 |
| Telestream Qualify (contract product) Telestream's blog describes AI lip-sync detection that reports shift and drift with confidence values and configurable thresholds, and says the capability is a Tech Preview. It gives no accuracy figures or formats. A February 2026 release, read only as a search summary, adds subtitle alignment and spoken-language checks. | Check files before delivery and search a media archive | Studio | Cloud or on your own servers | Not checked | 2026-10-08 |
| Moments Lab (contract product) Page describes cloud cataloguing with AI indexing, natural-language search through a Discovery Agent, a browser editor and publishing, with Adobe Premiere Pro, Amazon S3 and API integrations. Dalet is not mentioned on the page, so the design document claim of a Dalet integration is unconfirmed. Customer names are the page's own. | Check files before delivery and search a media archive | Studio | Cloud, upload required | Not checked | 2026-10-08 |
| ENCO aiTrack (contract product) ENCO's page says aiTrack lets stations build generative copy insertions with automated voice tracks into playlists well in advance of airtime, mixes live talent with AI content, and can use pre-set voice types or custom cloned voices. It says dynamic content insertion for its DAD automation system was still to come. The page names no voice engine, no languages and no human review step. Radio World reports a 2025 NAB Best of Show award and a 2024 preview that calls aiTrack a tool using generative AI models and synthetic speech engines. | Fill radio links with AI voice tracks and schedule by plain request | Studio | Not checked | Not checked | 2026-10-09 |
| RCS RCSAI in SelectorCloud (Announced, not yet released, contract product) Announced for debut at IBC2026. A trade article of 2 September 2026 says it lets programmers enter scheduling requests in natural language and is the first application on RCS Intelligence. No release date or price is stated, so check that it has shipped. RCS's own page was unreachable. | Fill radio links with AI voice tracks and schedule by plain request | Studio | Cloud, upload required | Not checked | Not yet |
| Futuri CoHostAI (contract product) Seen only in search summaries, which describe it as letting a human host and an AI personality interact on air in 2024. No page was read. | Fill radio links with AI voice tracks and schedule by plain request | Studio | Not checked | Not checked | Not yet |
| ElevenLabs The pricing page does not state the Free plan commercial terms. Its Starter list begins with Everything in free, plus, and then lists a commercial licence, which suggests Free has none. Not confirmed. | Make a synthetic voice or narration | Starter | Cloud, upload required | Not checked | 2026-10-08 |
| Respeecher (contract product) Page lists film, TV, games, broadcasting and other uses, and a Pro Tools plug-in. Its consent and voice-origin statements are the company's own and were not independently checked. No 2026 studio deal reporting found. | Make a synthetic voice or narration | Studio | Cloud, upload required | Not checked | 2026-10-08 |
| Upscayl Images only, not video. Use it on frames you extract. The README says it cannot de-blur. | Upscale and restore footage | Free | Runs on your computer | A Vulkan-compatible GPU is required. Most integrated graphics do not work. Windows 10 or macOS 12 or newer. | 2026-10-08 |
| TensorPix Browser tool. The page says processing happens in the cloud on remote GPUs and that local installation is not supported. It states a free allowance of about 5 minutes of video. | Upscale and restore footage | Starter | Cloud, upload required | Not checked | 2026-10-08 |
| Topaz Video AI Local rendering keeps files on your computer, per the page. Cloud rendering is also offered. | Upscale and restore footage, Restore and remaster damaged film and video | Pro | Runs on your computer | Mac and Windows. A legacy requirements page lists a 6 GB NVIDIA, AMD or Intel Arc A750 GPU minimum on Windows and 8 GB recommended, and 16 GB of RAM minimum. Current versions might differ. | 2026-10-08 |
| Telestream Vantage Super Resolution (Announced, not yet released, contract product) Announced 2 September 2026. Preview at IBC 2026, release planned Q4 2026, first on Vantage Lightspeed VLS 105, 205 and 305 servers with NVIDIA Blackwell GPUs. Not yet released. Re-check in 2027. | Upscale and restore footage | Studio | Runs on your computer | Not checked | Not yet |
| Krita AI Diffusion Cloud generation is also offered. The page gives no commercial-use statement for the models, so check each model licence before commercial work. | Fill, extend or remove in a photo | Free | Runs on your computer | At least 6 GB of NVIDIA graphics memory recommended. Without it generation is slow or fails. | 2026-10-08 |
| Adobe Photoshop Generative Fill | Fill, extend or remove in a photo | Pro | Cloud, upload required | Not checked | Not yet |
| Wan 2.2 in ComfyUI Apache 2.0 licence is stated to permit commercial use. The design document figure of about 8 GB was not found on this page and stays unverified. | Generate video | Free | Runs on your computer | Not stated on the ComfyUI announcement. The 5B variant is described as VRAM-optimised. | 2026-10-08 |
| Kling Page names Kling 4.0 for video and Kling Image 3.0. It states no free credits or commercial-use terms. Read the Terms of Use before commercial work. | Generate video | Starter | Not checked | Not checked | 2026-10-08 |
| Google Veo Veo 3.1 is the current version on the page. It generates audio with video, and output is watermarked with SynthID. Available through the Gemini app, Flow and the Gemini API. | Generate video | Pro | Cloud, upload required | Not checked | 2026-10-08 |
| Runway Page calls Runway Creative a cloud-based platform and names Gen-4.5 as its video model. "Try free" appears with no terms given. | Generate video | Pro | Cloud, upload required | Not checked | 2026-10-08 |
| Moonvalley Marey (contract product) Page says Marey is "trained on licensed data" and "commercially safe" without naming sources. The page makes no indemnity statement. One trade report says indemnity covers its data contractors only, which we did not read in full. | Generate video | Studio | Cloud, upload required | Not checked | 2026-10-08 |
Where AI is mature, and where the hype outpaces the result
This section carries over the editorial view of the guide this hub replaced. It is judgement, not a test result. Transcription on clean audio is the most dependable job. It feeds captions, searchable transcripts and text-based editing, where you cut speech by editing the words. Cleanup of poor recordings is useful, with a real cost: a tool set too hard damages audio that was already clean. Clip detection for cutting long video into short pieces saves time and still needs a person to review each cut.
The hype is loudest in three places. Video generated from text prompts is improving, and it remains unreliable when shots must match each other or a brand. Upscaling gives marginal gains on footage that was already decent, and it earns its place mainly on low-resolution archive material. Platform announcements are roadmap until the feature appears in your own upload screen. Platform auto-dubbing is an example: reports we read disagree on dates and on which creators have it, so check your own account. Our guide to dubbing workflows covers the human, AI and hybrid routes.
For one worked example that strings several of these jobs together, read A Practical AI-Assisted Edit. On provenance, read Content Credentials (C2PA).
Before you use AI on a job
Consent, labelling and ownership decide what you are allowed to publish. Read these pages first: AI Voice Cloning: Consent, Contracts, and Disclosure, When to Label Content as AI-Made: Platform Rules and Audience Trust and Protecting Your Work in the AI Era: Scraping, Impersonation, and Takedowns. Union agreements, European Union labelling rules and music licensing disputes also apply to professional work. We have read the union's own summary of the tentative 2026 SAG-AFTRA TV/Theatrical agreement, and our guide to it follows it. We have not read the full contract, the European Union labelling rules or music licensing court records, so this page does not summarise them.
How we choose
We pick the job first and the tool second. Where reviews disagree on a leader, the job record says so instead of naming a winner. Studio and contract products such as broadcast encoders are marked as contract products because we cannot test them. Tool rows are due for review every three months. To report an error, use the contact page. Corrections are checked like any other claim.
What changed
2026-10-08: first version, with 17 jobs and 48 tools. See the changelog for the site history.
Sources
Checked 2026-10-08. We opened the maker's own page for 39 of 48 tools. The other 9 are listed from trade press, vendor blogs and reviews: Adobe Enhance Speech, Avid Pro Tools Speech-to-Text, Avid Media Composer, Autodesk Flow Studio, AI-Media LEXI, RCS RCSAI in SelectorCloud, Futuri CoHostAI, Telestream Vantage Super Resolution, Adobe Photoshop Generative Fill. Their makers' pages blocked our request, returned no product text, or did not mention the product. We tested none of the tools. Where a figure comes from a page we did not read, the row says "Not yet" under "Checked against maker". We record no prices, so the tier for each tool is our estimate from those sources and is labelled Estimated.
- Audacity OpenVINO AI effects. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Descript Studio Sound. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Antinode Clear (formerly Supertone Clear). The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Waves Clarity Vx Pro. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Accentize dxRevive Pro. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- iZotope RX. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- CEDAR DNS. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- NVIDIA Broadcast. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Krisp. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Ultimate Vocal Remover. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Moises. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- LALAL.AI. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Steinberg SpectraLayers. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- AudioShake. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- OpenAI Whisper. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- NUGEN Halo Upmix. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Adobe Premiere Pro. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- DaVinci Resolve Studio. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Foundry Nuke CopyCat. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Beeble. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- MTI Film DRS Nova. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Phoenix restoration software (filmworkz.com). The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Verbit (VITAC Captivate). The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- CAMB.AI. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Interra Systems BATON. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Telestream Qualify. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Moments Lab. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- ENCO aiTrack. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- ElevenLabs. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Respeecher. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Upscayl. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- TensorPix. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Topaz Video AI. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Krita AI Diffusion. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Wan 2.2 in ComfyUI. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Kling. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Google Veo. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Runway. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
- Moonvalley Marey. The maker's own page. Read for what the tool does, and for licence, platform and hardware where the page states them.
Secondary sources behind the unchecked rows
These identify products and report claims. They are not the product records. Several are written by vendors or carry affiliate links, so no tool is ranked from them alone.
- aap.com.au on LEXI encoders at NAB 2026, search summary only
- advanced-television.com
- autodesk.com/products/flow-studio/overview, page returned 403
- avid.com/media-composer, page returned 403
- avid.com/press-room/2026/04 (Pro Tools 2026.4), page returned 403
- blog.adobe.com 2025-09-25 Photoshop Beta partner models
- diyai.io Adobe Podcast review
- dupple.com best AI audio enhancement tools
- investorrelations.ai-media.tv release, returned HTTP 429, not read
- kompozy.io photoshop generative fill model picker
- newsshooter.com/2026/08/27/avid-media-composer-2026-8, read 2026-10-08
- prosoundweb.com/avid-announces-latest-pro-tools-release, read 2026-10-08
- rcsworks.com/selectorcloud, returned HTTP 403, not read
- redtech.pro 2026-09-02 on the IBC2026 debut, read
- search summaries of Radio Ink, Inside Radio and Business Wire coverage, pages returned 403 or 429, not read
- search summary of autodesk.com Flow Studio product-details pages, not opened
- sportsvideo.org 2026-04-15 on LEXI Text and Voice Encoders, read
- sportsvideo.org/2026/02/26
One correction to our own earlier notes
An early draft said iZotope's parent group was sold to investors with effect from 1 July 2026. MusicTech, 2 July 2026, instead reports that Boris FX acquired iZotope, gives no exact sale date, and says licences and subscriptions stay active. It is the only record we read, so treat the detail as reported.
Translate this page
- Español
- 简体中文
- हिन्दी
- العربية
- Português
- Français
- Deutsch
- 日本語
- Русский
- Bahasa Indonesia
- 한국어
- Italiano
- Türkçe
- Tiếng Việt
- Polski
- Nederlands
Machine translation provided by Google Translate, on Google’s servers. We do not check these translations and they will get technical terms wrong. The English page is the authoritative one. Following a link sends this page’s address to Google. Your browser may also offer to translate this page itself, which keeps the request on your device.