AI models

49 AI models, explained. With real samples.

Every model in the studio has its own page: what it is, results we made with it, where it is strong and where it falls short. 49 models from 18 companies, on one account.

10 free tokens to start

Image models

Models that draw pictures from words, change a picture you upload, or fix one.

Made with Nano Banana 2Google · ImageNano Banana 2Google's all-rounder. It follows long instructions closely, writes words inside pictures well and changes only what you ask.from 3 tokensSee samplesMade with Nano Banana ProGoogle · ImageNano Banana ProGoogle's Pro model: the finest detail and the most reliable small text here. Slower, and twice the price of Nano Banana 2.from 6 tokensSee samplesMade with Seedream 4.5ByteDance · ImageSeedream 4.5Rich colour and strong composition for two tokens, and the most faithful to a real face in our tests.from 2 tokensSee samplesMade with Seedream 5 ProByteDance · ImageSeedream 5 ProByteDance's top model: clean 2K pictures with the finest detail. Slower than the others.from 4 tokensSee samplesMade with FLUX.2 proBlack Forest Labs · ImageFLUX.2 proPhotographs with natural light and skin, made fast. It keeps people and products true while the scene around them changes.from 2 tokensSee samplesMade with FLUX.2 maxBlack Forest Labs · ImageFLUX.2 maxThe largest FLUX model: the most detail and the closest to the prompt in the family.from 4 tokensSee samplesMade with FLUX 1.1 proBlack Forest Labs · ImageFLUX 1.1 proFast, polished and reliable. A safe choice for marketing pictures.from 2 tokensSee samplesMade with FLUX.2 devBlack Forest Labs · ImageFLUX.2 devThe open FLUX.2 model: close to pro quality for one token.from 1 tokenSee samplesMade with FLUX.1 devBlack Forest Labs · ImageFLUX.1 devThe classic open FLUX model. Dependable and inexpensive.from 1 tokenSee samplesMade with FLUX.1 schnellBlack Forest Labs · ImageFLUX.1 schnellThe fastest model here: a draft in about two seconds.from 1 tokenSee samplesMade with Juggernaut FLUXRunDiffusion · ImageJuggernaut FLUXA FLUX fine-tune for lifelike subjects and cinematic photographs, for one token.from 1 tokenSee samplesMade with Wan 2.6Alibaba · ImageWan 2.6Vivid, stylised pictures with strong lighting.from 2 tokensSee samplesMade with P-ImagePruna · ImageP-ImageVery fast and very cheap. Handy for trying ideas.from 1 tokenSee samplesOpenAI · ImageGPT Image 2.6The top-ranked picture model. It follows long, exact instructions and keeps a person the same from up to four reference pictures.from 2 tokensRead moreMade with Qwen ImageAlibaba · ImageQwen ImageOpen-source pictures for a film at about a third of the price: one model draws from words, a second keeps people and places from references.from 1 tokenSee samplesMade with Wan 2.7 EditAlibaba · ImageWan 2.7 EditBold, stylised edits and strong lighting changes for two tokens.from 2 tokensSee samplesMade with P-Image EditPruna · ImageP-Image EditThe quickest and cheapest editor, for simple changes.from 1 tokenSee samplesMade with Clarity CrystalClarityAI · ImageClarity CrystalFaithful enlargement: faces, products and text stay as they are, only sharper.from 5 tokensSee samplesClarityAI · ImageClarity CreativeAn upscaler that redraws fine detail. Lovely on photos, landscapes and art; wrong for anything with text.from 5 tokensRead moreMade with Bria RMBG 2.0Bria · ImageBria RMBG 2.0Clean cut-outs with soft edges and hair, as a transparent PNG.from 3 tokensSee samplesMade with Bria Replace BackgroundBria · ImageBria Replace BackgroundKeeps the subject exactly and builds a new setting around it from your description.from 3 tokensSee samplesMade with Bria ExpandBria · ImageBria ExpandContinues a picture beyond its edges, so a square photo becomes a banner or a story.from 5 tokensSee samplesMade with Bria EnhanceBria · ImageBria EnhanceRemoves noise, blur and compression in one pass, without changing the picture.from 3 tokensSee samplesMade with Bria Blur BackgroundBria · ImageBria Blur BackgroundFinds the subject and softens everything behind it, like a portrait lens.from 3 tokensSee samples

Video models

Models that make a moving clip from words or from a first frame.

Made with Seedance 2.5ByteDance · VideoSeedance 2.5The top-ranked video model: cinematic motion, sound and lip-synced speech, starting from the shot's first frame. The most expensive.16 tokens a secondSee samplesMade with Seedance 2.0ByteDance · VideoSeedance 2.0Cinematic camera work and sound from a written shot. It refuses realistic people, so it is for scenery, products and animation.6 tokens a secondSee samplesMade with Wan 3.0Alibaba · VideoWan 3.0Natural movement with sound and speech, up to 30 seconds, from the shot's first frame. The High quality setting.4 tokens a secondSee samplesKuaishou · VideoKling 4.0Kling's latest picture quality and motion, up to 30 seconds. It makes no sound of its own.6 tokens a secondRead moreKuaishou · VideoKling 3.0 ProKling with sound: natural acting and speech in English, from the shot's first frame.6 tokens a secondRead moreKuaishou · VideoKling 2.6 ProThe earlier Kling: natural acting and motion with sound, in 5 or 10 second clips.5 tokens a secondRead moreKuaishou · VideoKling AvatarFor talking shots: the face in the first frame speaks a voice line that was recorded first.2 tokens a secondRead moreGoogle · VideoVeo 3.1Google's best performances, speech and sound. It is made from the written shot, so faces can change between shots.14 tokens a secondRead moreGoogle · VideoVeo 3.1 FastThe quicker, cheaper Veo: characters speak their lines with sound, made from the written shot.5 tokens a secondRead moreMade with LTX 2.3 FastLightricks · VideoLTX 2.3 FastFast 720p video with sound for one token a second. The studio's default, and the best value for long films.from 6 tokensSee samplesMade with LTX 2.5 FastLightricks · VideoLTX 2.5 FastThe newer LTX: sharper detail and steadier motion than 2.3, with sound, up to 20 seconds.from 18 tokensSee samplesMade with LTX 2.5 ProLightricks · VideoLTX 2.5 ProLTX's highest fidelity, with sound. For the shots that matter, in clips of 6 to 10 seconds.from 24 tokensSee samplesMade with LTX-2.5 open sourceLightricks · VideoLTX-2.5 open sourceThe open-source LTX-2.5 running on our own graphics cards: a sharp picture with sound at the lowest price in the studio.0.6 tokens a secondSee samplesMade with Wan 2.2 open sourceAlibaba · VideoWan 2.2 open sourceThe open-source Wan 2.2 on our own GPU: the most natural movement of the open models, for shots where people walk, turn or work.1 token a secondSee samplesMade with Wan 2.7Alibaba · VideoWan 2.7Makes a character speak in time with a recorded voice line, or keeps the cast the same from reference pictures.4 tokens a secondSee samplesMade with Pixverse 6Pixverse · VideoPixverse 6Smooth, stylised motion. Popular for social clips.from 8 tokensSee samplesMade with P-VideoPruna · VideoP-VideoThe cheapest clip here. Good for checking an idea before a final take.from 4 tokensSee samplesMade with P-Video AvatarPruna · VideoP-Video AvatarCheap talking shots: a face moves its mouth in time with its voice line.1 token a secondSee samplesMiniMax · VideoHailuo 02Lively, realistic motion at a low price. No sound: add a voice-over or music.2 tokens a secondRead more

The same prompt on 13 models

One sentence, sent to every picture model with no changes. This is the quickest way to see how differently they read the same words.

Asked forA red fox wearing a tiny knitted crimson scarf sitting at a cafe table in the snow, warm window light behind, cinematic photograph, shallow depth of field

The same edit on 6 editors

The original shoe is red. Each editor was given the same picture and the same instruction. Look at how close each blue is to cobalt, and at what else changed.

Asked forChange the colour of the shoe to bright cobalt blue. Keep everything else exactly the same.

The same line on 4 voices

One sentence read by each voice model. Press play to hear the difference.

The lineMake any image better, in seconds. Sharpen it, cut it out, widen it, or put your logo on a real product.

How we test

Every picture, clip and voice on these pages was made by the model it is shown under, in this studio or in our own tests. Nothing comes from a maker's showreel, and nothing is retouched.

Strengths and weaknesses are what we saw: the same prompt on every model, and the films and designs we have made. Where we have not run a model ourselves yet, its page says so and shows no sample.

Where a public ranking is quoted, it is from Artificial Analysis, read on 20 November 2026. Rankings move as new models arrive, so treat them as a guide.

Looking for a job to do, not a model? See the tools.

Pick a model. Make something.

Create an account, pick a pack from $6, and drop in a picture.

10 free tokens to start