AIVideoFreeEvery model we run, and what each one is actually for
Twenty-three models, one line each on the job it wins. The useful question is never "which is best" — it's "which is best at this", and every page below answers that honestly, including where the model loses.
Video models
Free daily drafts run on Seedance Fast; each page below tells you when it's worth spending credits and when it isn't.
Veo
The realism benchmark every summer flagship measured itself against. Footage that passes as real, with ambience and dialogue in the same pass.
Seedance 2.5
One unbroken 30-second take, 50 reference files, and a soundtrack that can drive the pacing and the cut.
Kling 3.0
Director Mode composes up to six shots — cuts included — in a single generation, at up to 4K with multilingual lip-sync.
Grok Imagine 1.5
Your still is treated as literally frame one. Took the image-to-video arena crown on that, and renders in seconds.
MiniMax H3
The cheapest native 2K, with stereo audio, dialogue and a 12-file omni-reference intake.
Wan 3.0
The only model here that reads documents — hand it a deck, a PDF or a spreadsheet and it builds the video from the contents.
FLUX 3
Pin up to ten keyframes and get synced audio in the same pass. Stop-motion to cinematic, without the glossy default look.
Kling O3
Reference-locked identity across shots, multi-character scenes, and conversational edits applied to finished clips.
Luma Ray 3.2
Sixteen keyframes per clip and HDR with 16-bit EXR — the only output here a colourist can grade like footage.
Seedance 2.0
The arena champion of early 2026, now the value tier: native audio and proven quality at the family's lowest paid rate.
Hailuo 2.3
Human motion that reads as human, at the Hailuo line's lowest rate. The version most existing Hailuo reviews describe.
Wan 2.7
The stable Wan: four mature workflows on production capacity, no invite wall, schedulable on a client deadline.
Image models
Same rule: draft cheap, finish on the model whose page explains why it costs more.
GPT Image 2.5
Half the previous latency, warmer light, and edits that hold their subject turn after turn. The newest model on this site.
Nano Banana 2
Google's own default: Pro-level output at Flash speed, 512px to 4K, with web-image grounding before it draws.
Seedream 5.0 Pro
The only flagship that hands you 10+ editable transparent layers instead of a flat file, plus point-and-lasso editing.
FLUX.3
Multi-reference brand control and exact hex colours, with an Apache-licensed tier you can self-host.
Ideogram Character
Typography's benchmark, now open-weight: dense multi-line text, JSON layout control, 32 scripts.
Grok Imagine 2
The only model that outputs true editable SVG — logos, icon systems and illustration that scale losslessly.
GPT Image 2
April's arena record-holder by the largest margin ever recorded. Thinking mode and print-ready text.
Nano Banana Pro
Reasoning-backed accuracy for infographics, diagrams and layouts where every number has to land.
Seedream 5.0 Lite
The reasoning model at sketchpad prices: fourteen references, high-volume exploration.
Qwen-Image 3.0
Prompts up to 4,500 tokens and text legible at ten pixels. Documents and dense layouts at utility pricing.
Tools and editors
Jobs rather than models — the engine is picked for you, or routed on the page.
Remove, replace, restyle — ten models behind one box.
Fix footage you already have, with a free preview every time.
Which public figure you resemble, and why.