Image generation models
Stills, and the models that edit them.
- Catalogued here
- 20 models
- Open weights
- 5
- With an API
- 18
What to know about image generation
Stills, and the models that edit them. Resolution matters less than control here: whether you can hold a face across generations, put legible text inside the frame, and get the same composition twice. Licensing is more varied than in video, and several of these models split their terms by variant, so the weights you downloaded decide what you may do with the output.
Duration and resolution, side by side
Only the models whose vendor publishes a figure appear here. "Not established" means no page we could retrieve stated one.
| Model | Duration | Resolution | Weights | Commercial use |
|---|---|---|---|---|
| Adobe Firefly Image Model 5 | not established | native 4 MP without upscaling; Upscale API adds 2x / 4x / 6x | Not released | Permitted subject to conditions |
| FLUX.2 [dev] | not established | 4 MP for image editing (FLUX.2 family, per BFL launch post); no hard ceiling published | Downloadable | The licence forbids commercial use |
| FLUX.2 [flex] | not established | 4 MP for image editing (FLUX.2 family, per BFL launch post); exact max width/height not published | Not released | Vendor states commercial use is permitted |
| FLUX.2 [max] | not established | 4 MP for image editing (FLUX.2 family, per BFL launch post); exact max width/height not published | Not released | Vendor states commercial use is permitted |
| FLUX.2 [pro] | not established | 4 MP for image editing (FLUX.2 family, per BFL launch post); exact max width/height not published | Not released | Vendor states commercial use is permitted |
| Ideogram 4.0 | not established | 2048 px on the long side (native resolutions 256-2048 in multiples of 16) | Downloadable | The licence forbids commercial use |
| Imagen 4 (Fast / Standard / Ultra) | not established | 2K (1K default; Ultra and Standard support 1K and 2K) | Not released | Not established |
| Nano Banana 2 (Gemini 3.1 Flash Image) | not established | 4K (0.5K / 1K / 2K / 4K output options) | Not released | Not established |
| Nano Banana Pro (Gemini 3 Pro Image) | not established | 4K (1K / 2K / 4K output options) | Not released | Not established |
| Recraft V4.1 Vector | not established | resolution-independent (SVG output) | Not released | Permitted subject to conditions |
| Stable Diffusion 3.5 Large | not established | ~1 megapixel (Stability describes SD3.5 Large as ideal for professional use at 1 megapixel resolution) | Downloadable | Permitted subject to conditions |
All 20
-
Adobe Firefly Image Model 5
Adobe's current Firefly image model, announced in public beta at Adobe MAX on 28 October 2025, generating at native 4 MP resolution without an upscale pass. It powers a conversational 'Promp
-
FLUX.1 Kontext [pro] / [max]
Instruction-driven image editing models from the FLUX.1 generation (2025), still exposed as live /flux-kontext-pro and /flux-kontext-max endpoints. They take an input image plus a text instr
-
FLUX.2 [dev]
32-billion-parameter open-weights checkpoint that does text-to-image plus single- and multi-reference editing in one model, released 25 November 2025 on Hugging Face. Runs on consumer GPUs w
-
FLUX.2 [flex]
Hosted FLUX.2 variant that exposes sampling steps and guidance scale as API parameters, letting callers trade quality and prompt adherence against speed and cost. BFL highlights it specifica
-
FLUX.2 [klein] (4B / 9B)
Size-distilled open-weights members of the FLUX.2 family, released January 2026 in 4B and 9B sizes plus 'Base' checkpoints intended for fine-tuning and LoRA training. They keep text-to-image
-
FLUX.2 [max]
Top-tier hosted model in the FLUX.2 family, exposed as the /flux-2-max endpoint on the BFL API. Handles text-to-image plus single- and multi-reference editing in one model, with the FLUX.2 l
-
FLUX.2 [pro]
The main hosted FLUX.2 model, launched 25 November 2025, offered as both a rolling preview endpoint and a pinned snapshot for reproducible workflows. Combines text-to-image with single- and
-
GPT Image 2
OpenAI's current image model (default snapshot gpt-image-2-2026-04-21), served on both /v1/images/generations and /v1/images/edits with documented inpainting support. It takes text and image
-
Ideogram 3.0
Previous Ideogram flagship, still fully priced and served on the API, and the version that carries the Character reference feature: one uploaded reference image keeps a subject's face and ke
-
Ideogram 4.0
Ideogram's current flagship, a 9.3B-parameter diffusion transformer released 3 June 2026 with both API access and downloadable weights. Its distinguishing feature is layout-first generation:
-
Imagen 4 (Fast / Standard / Ultra)
Google's standalone dedicated text-to-image family (imagen-4.0-fast-generate-001, imagen-4.0-generate-001, imagen-4.0-ultra-generate-001), offering 1K or 2K output, five fixed aspect ratios,
-
Krea 2
Krea's first from-scratch foundation image model, a 12-billion-parameter text-to-image diffusion model released 22 June 2026 with weights published on Hugging Face. It ships as Krea 2 Raw (b
-
Luma Photon 1 / Photon Flash 1
Luma's image models (photon-1 default, photon-flash-1 for speed) on the same API platform as its Dream Machine video models. Both expose the full control set: image reference with up to 4 im
-
Midjourney V8.2
Midjourney's current default image model, released 24 July 2026 as an aesthetics and reliability update over V8.1. Vendor notes emphasise bolder, more sophisticated output, far fewer low-qua
-
Nano Banana 2 (Gemini 3.1 Flash Image)
Google's workhorse image model, model ID gemini-3.1-flash-image, described in the docs as the most versatile generalist of the Nano Banana line and the default image generator in the Gemini
-
Nano Banana Pro (Gemini 3 Pro Image)
Google's premium image generation and editing model, model ID gemini-3-pro-image, built on Gemini 3 Pro and announced 20 November 2025. It generates at 1K, 2K and 4K, renders legible multili
-
Recraft V4.1 (raster)
Recraft's current raster model, released 14 May 2026, shipped in eight variants across Standard/Pro (higher resolution and finer detail for print) and Utility/Utility Pro (deliberately plain
-
Recraft V4.1 Vector
Recraft's vector generation path, which outputs genuinely editable SVG files rather than raster images that need tracing. Available as V4.1 Vector, V4.1 Vector Pro, and Utility vector varian
-
Stable Diffusion 3.5 Large
Stability AI's 8-billion-parameter open-weights flagship, a multimodal diffusion transformer aimed at roughly 1-megapixel output. It ships alongside Large Turbo (fewer steps) and Medium (2.5
-
Stable Image Ultra
Stability's highest-quality hosted text-to-image service on its developer platform, positioned above Stable Image Core and updated to run on the Stable Diffusion 3.5 Large base. It is a mana