Choose the right model for every image and video
Halftone AI runs on a curated set of image and video models from Black Forest Labs, Google, ByteDance, Kuaishou, MiniMax, OpenAI, Recraft, and xAI. Each one has real strengths and real limits — this page breaks down what each model actually does, so you can pick the right one before you spend a credit.
All models
Flux 2 Pro
Photorealistic detail with reference-image consistency.
Flux 2 Max
BFL's flagship, grounded in live web information.
Ideogram v3
The most reliable model here for legible text in images.
Recraft v4
The only model here built for brand and vector work.
Recraft v4.1
The sharper, budget-friendly Recraft refresh.
Recraft v4.1 Pro
Higher-resolution design work for premium and print output.
Seedream 4.5
Strong spatial reasoning for multi-subject scenes.
Nano Banana Pro
Conversational editing that remembers your character.
GPT Image 2
Dense, multilingual text with precise multi-image edits.
Wan 2.7 Pro
Fast, affordable photorealism for everyday generation.
Grok Imagine Image
Best-in-class instruction following for creative direction.
Grok Imagine Image 2.0
The next-generation Grok Imagine — higher fidelity, genuine 2K.
Muse Image
Agentic generation and editing, the cheapest model in the catalog.
Kling Image O3
4K Omni generation with strong subject consistency.
Qwen Image 3.0
Balanced quality, layout fidelity, and speed for everyday generation.
Qwen Image 3.0 Pro
High-control generation for dense layouts and small, multilingual text.
Kling v3
Fluid, believable human motion for action and dance.
Veo 3.1
True 4K cinematic video with native, synced audio.
Seedance 2.0
Multi-shot cinematic sequences with native audio.
Seedance 2.5
Native 30-second clips with in-place editing.
Seedance 2.0 Mini
Compact Seedance, for fast and light multimodal runs.
Hailuo 2.3
The most expressive character motion in this lineup.
MiniMax H3
Native synced audio, multi-reference, and continuation workflows.
MiniMax H3 Max
High-throughput MiniMax H3, with first-to-last-frame control.
MiniMax H3 Fast
The fastest, cheapest MiniMax H3 tier.
Wan 2.7
Plans the shot before generating it — affordable multi-shot video.
Wan 3.0
Native 30-second clips with a much larger reference cast.
Wan 3.0 Prime
Wan 3.0, with lower latency.
Grok Imagine Video
Precise camera direction, strongest as image-to-video.
Grok Imagine Video 1.5
Longer, higher-fidelity xAI video with multi-speaker dialogue.
Seedream 5 Pro
ByteDance’s flagship image model, tiered by output resolution.
Veo 3.1 Lite
The cheapest way onto Veo, with native audio intact.
Kling o3 Pro
Long-form Kling with native audio, up to 15 seconds.
Kling v2.5 Turbo Pro
Fast, economical Kling for silent text-to-video.
Kling O1
Image-to-video only — animates a still you supply.
Kling v3 Turbo
Faster, cheaper Kling v3 motion for quick iteration.
Kling v3 4K
Native 4K output at a flat per-second rate.
Flux 3 Video
Multimodal video with native synchronized audio.
LTX 2.5 Pro
Production-quality video with native audio and named camera moves.
HappyHorse 1.1
Multi-character video with strong reference consistency.
When to use which model
- Cinematic video with native audio
- Text-heavy graphics — posters, signage, UI
- Brand and vector assets
- Conversational, multi-turn photo editing
- Multi-shot narrative video on a budget
- Precise camera direction from a still image
Need to remove a background, inpaint, upscale, or relight an existing photo instead of generating a new one? See editing & utility tools.
Compare speed, realism, creativity, and control
Image models
| Model | Speed | Realism | Creativity | Control | Best input | Ideal output |
|---|---|---|---|---|---|---|
| Flux 2 Pro | text, image-ref | Photorealistic detail with reference-image consistency. | ||||
| Flux 2 Max | text, image-ref | BFL's flagship, grounded in live web information. | ||||
| Ideogram v3 | text | The most reliable model here for legible text in images. | ||||
| Recraft v4 | text | The only model here built for brand and vector work. | ||||
| Recraft v4.1 | text | The sharper, budget-friendly Recraft refresh. | ||||
| Recraft v4.1 Pro | text | Higher-resolution design work for premium and print output. | ||||
| Seedream 4.5 | text, image-ref | Strong spatial reasoning for multi-subject scenes. | ||||
| Nano Banana Pro | text, image-ref | Conversational editing that remembers your character. | ||||
| GPT Image 2 | text, image-ref | Dense, multilingual text with precise multi-image edits. | ||||
| Wan 2.7 Pro | text, image-ref | Fast, affordable photorealism for everyday generation. | ||||
| Grok Imagine Image | text, image-ref | Best-in-class instruction following for creative direction. | ||||
| Grok Imagine Image 2.0 | text, image-ref | The next-generation Grok Imagine — higher fidelity, genuine 2K. | ||||
| Muse Image | text, image-ref | Agentic generation and editing, the cheapest model in the catalog. | ||||
| Kling Image O3 | text, image-ref | 4K Omni generation with strong subject consistency. | ||||
| Qwen Image 3.0 | text, image-ref | Balanced quality, layout fidelity, and speed for everyday generation. | ||||
| Qwen Image 3.0 Pro | text, image-ref | High-control generation for dense layouts and small, multilingual text. | ||||
| Seedream 5 Pro | text, image-ref | ByteDance’s flagship image model, tiered by output resolution. |
Video models
| Model | Speed | Realism | Creativity | Control | Best input | Ideal output |
|---|---|---|---|---|---|---|
| Kling v3 | text, image-ref, video-ref | Fluid, believable human motion for action and dance. | ||||
| Veo 3.1 | text, image-ref | True 4K cinematic video with native, synced audio. | ||||
| Seedance 2.0 | text, image-ref, video-ref, audio-ref | Multi-shot cinematic sequences with native audio. | ||||
| Seedance 2.5 | text, image-ref, video-ref, audio-ref | Native 30-second clips with in-place editing. | ||||
| Seedance 2.0 Mini | text, image-ref, video-ref, audio-ref | Compact Seedance, for fast and light multimodal runs. | ||||
| Hailuo 2.3 | text, image-ref | The most expressive character motion in this lineup. | ||||
| MiniMax H3 | text, image-ref, video-ref, audio-ref | Native synced audio, multi-reference, and continuation workflows. | ||||
| MiniMax H3 Max | text, image-ref, video-ref, audio-ref | High-throughput MiniMax H3, with first-to-last-frame control. | ||||
| MiniMax H3 Fast | text, image-ref, video-ref, audio-ref | The fastest, cheapest MiniMax H3 tier. | ||||
| Wan 2.7 | text, image-ref | Plans the shot before generating it — affordable multi-shot video. | ||||
| Wan 3.0 | text, image-ref, video-ref, audio-ref | Native 30-second clips with a much larger reference cast. | ||||
| Wan 3.0 Prime | text, image-ref, video-ref, audio-ref | Wan 3.0, with lower latency. | ||||
| Grok Imagine Video | text, image-ref | Precise camera direction, strongest as image-to-video. | ||||
| Grok Imagine Video 1.5 | text, image-ref | Longer, higher-fidelity xAI video with multi-speaker dialogue. | ||||
| Veo 3.1 Lite | text, image-ref | The cheapest way onto Veo, with native audio intact. | ||||
| Kling o3 Pro | text, image-ref | Long-form Kling with native audio, up to 15 seconds. | ||||
| Kling v2.5 Turbo Pro | text | Fast, economical Kling for silent text-to-video. | ||||
| Kling O1 | image-ref | Image-to-video only — animates a still you supply. | ||||
| Kling v3 Turbo | text, image-ref | Faster, cheaper Kling v3 motion for quick iteration. | ||||
| Kling v3 4K | text, image-ref | Native 4K output at a flat per-second rate. | ||||
| Flux 3 Video | text, image-ref, video-ref | Multimodal video with native synchronized audio. | ||||
| LTX 2.5 Pro | text, image-ref | Production-quality video with native audio and named camera moves. | ||||
| HappyHorse 1.1 | text, image-ref | Multi-character video with strong reference consistency. |