Modèles

Tous les modèles, un seul pool de crédits.

Vingt modèles IA vidéo et image sur un seul pool de crédits : prix réels par clip, résolution et durée maximales, et lesquels des dix tiennent un personnage sur tout un court métrage.

Modèles vidéo11

popular

Kling V3.0Kling 3.0 by Kuaishou delivers state-of-the-art video generation with exceptional prompt adherence, character consistency, and physics simulation. It supports text-to-video, image-to-video, and advanced camera controls with output up to 1080p at 24fps.Voir le modèle

Wan 3.0Wan 3.0 by Alibaba animates a start image you provide, for anything from 2 to 15 seconds, and makes sound with the picture. It runs on Reelfo as a managed API call — nothing to install, no hardware to rent. A single-clip model: it starts from one frame and does not hold a cast across shots.Voir le modèle

new

Seedance 2.0Seedance by ByteDance specializes in human body motion, dance choreography, and expressive character animation. It generates videos with natural body physics, complex multi-joint movement, and coordinated dance sequences.Voir le modèle

new

Seedance 2.5Seedance 2.5 by ByteDance is the larger sibling of Seedance 2.0: more reference images a shot, audio references that can steer a voice or a beat, and clips that run up to thirty seconds. It makes its sound with the picture, and it runs on Reelfo as a managed API call — nothing to install, no hardware to rent.Voir le modèle

Hailuo 2.3Hailuo by MiniMax produces highly aesthetic, stylized video content with a distinctive cinematic look. Known for beautiful color grading, smooth motion, and artistic flair that makes every output feel professionally directed.Voir le modèle

popular

Veo 3.1Veo 3.1 is Google DeepMind's video model, and the only one in this catalog that treats audio as part of the generation rather than something you add afterwards. It also takes reference images for subject consistency, which is what lets it sit inside a reel instead of only producing standalone clips.Voir le modèle

Grok Imagine VideoGrok Imagine Video is xAI's video model. It accepts up to seven reference images — more than most models here — while costing less per clip than anything else we publish. For volume work where every shot does not need to be a hero shot, that combination is hard to argue with.Voir le modèle

new

Kling O3Kling O3 is Kuaishou's newer video line and a strict superset of Kling V3 at the same price. It takes the same Elements payload for character identity, adds a separate channel for style references, and can compose a multi-shot sequence from one request — and unlike V3 it is a reference-to-video endpoint, so it needs no start frame at all.Voir le modèle

PixVerse C1PixVerse C1 names each reference rather than numbering it — every image carries a short handle the prompt addresses directly, so a shot with several distinct subjects stays unambiguous about which is which. Seven references at a price near the bottom of the catalog.Voir le modèle

Vidu Q3Vidu Q3 generates video from reference images rather than from a single start frame. That distinction is the whole point in a reel: a start-frame model only knows what shot one looked like, while Vidu is handed the character sheet itself and can therefore keep that person recognisable in a shot it has never seen.Voir le modèle

HappyHorse 1.0HappyHorse 1.0 is a lip-sync-oriented video model. Most video models treat a talking head as a body that happens to move its mouth; HappyHorse treats the dialogue as the driver. For a reel with actual spoken lines, that difference is the whole shot.Voir le modèle

Modèles image09

Flux 2 ProFlux 2 Pro is Black Forest Labs' present flagship. Flux earned its reputation on the earlier dev and schnell weights, which is why most tutorials still point there — Flux 2 Pro is a different and considerably better model, and it is the one worth using for a storyboard frame you intend to keep.Voir le modèle

Grok ImageGrok Image is xAI's text-to-image model, available here on both of our providers. It sits at the cheap end of the catalog without being a draft-quality model, which makes it a reasonable default for boards you expect to iterate on rather than keep.Voir le modèle

Nano Banana ProNano Banana Pro is Google's gemini-3-pro-image, the strongest instruction-follower in the Nano Banana family. It is the model to reach for when a shot carries real staging — two characters, a specific camera height, a named light source — and a cheaper model would quietly drop half of it.Voir le modèle

Nano Banana 2Nano Banana 2 is Google's gemini-3.1-flash-image — newer than the original Nano Banana, cheaper than Pro. For most storyboard work it is the setting you leave it on: fast enough to re-render a board, accurate enough that you rarely regenerate a shot twice.Voir le modèle

Nano Banana 2 LiteNano Banana 2 Lite is Google's gemini-3.1-flash-lite-image: the fastest and least expensive member of the family. Its job in a reel workflow is the rough pass — block out ten shots, see which ones are working, then regenerate the keepers on a stronger model.Voir le modèle

Qwen Image 2.0Qwen Image 2.0 is Alibaba's image model, and the strongest option in this catalog for rendering text inside an image — particularly Chinese, Japanese and Korean, where most Western-trained models produce characters that look right and mean nothing. Title cards, signage, captions and posters are where it earns its place.Voir le modèle

Seedream 5.0 ProSeedream by ByteDance produces images with a distinctive artistic quality. Known for beautiful lighting, painterly textures, and creative compositions that go beyond photorealism into something more expressive and emotionally resonant.Voir le modèle

Nano BananaNano Banana is Google's gemini-2.5-flash-image, the model that made image-to-image character consistency mainstream. In Reelfo it does the job it is best at: you feed it one reference of your character and it returns that same character in a new pose, outfit or setting — which is exactly what a storyboard needs before any of it moves.Voir le modèle

GPT Image 2GPT Image by OpenAI combines world-class language understanding with image generation. It excels at following complex, nuanced prompts and producing images that precisely match detailed descriptions, including scenes with specific spatial relationships and interactions.Voir le modèle