Local image models
Create and edit images, and generate 3D.
Documented GPU memory
Lowest published path per model · GB, log scale · lower is better · 30 of 128 models
- Color = publisher
Source: publishers' documented requirements
Prompt alignment · GenEval
0–1 · higher is better · published results, not local measurements
- Stable Diffusion XL 1.00.55Stability AI
- Stable Diffusion 1.50.43Stability AI
Source: geneval-original-2023 · checked Oct 5, 2026
Qwen-Image: 4 GB (conditional path)
128 models · 30 with documented memory
Text to image: 88 models
Choose your machine and the charts and tables mark what fits in its memory.
Image models
| Qwen-Image-2.1 | 7.1 | No publisher documents a memory path for this model | Repository size not read for this model | Sep 2026 | Publisher terms |
|---|---|---|---|---|---|
| Cosmos3-Super-Text2Image-4Step | 64 | 640 | Repository size not read for this model | Jul 2026 | OpenMDW 1.1 |
| Ideogram 4.0 | 9.3 | 24 · conditional | 16.1 | Jun 2026 | Ideogram 4 Non-Commercial |
| Ideogram 4.0 (Quality) | 9.3 | 24 · conditional | 16.1 | Jun 2026 | Ideogram 4 Non-Commercial |
| Cosmos3-Super-Text2Image | 64 | 640 | 132.5 | May 2026 | OpenMDW 1.1 |
| Cosmos3-Super-Text2Image (agentic) | 64 | 640 | 132.5 | May 2026 | OpenMDW 1.1 |
| HiDream-O1-Image | 8 | No publisher documents a memory path for this model | 35.2 | May 2026 | MIT |
| HiDream-O1-Image-Dev | 8 | No publisher documents a memory path for this model | 35.2 | May 2026 | MIT |
| ERNIE Image | 8 | 24 | 31.6 | Apr 2026 | Apache 2.0 |
| ERNIE Image Turbo | 8.03 | 24 | 31.6 | Apr 2026 | Apache 2.0 |
| HunyuanImage 3.0 Instruct (Fal) | 80 | 640 | 168.7 | Jan 2026 | Tencent Hunyuan Community |
| FLUX.2 [klein] 4B | 4 | 13 | 23.7 | Jan 2026 | Apache 2.0 |
| FLUX.2 [klein] 9B | 9 | 29 | 52.9 | Jan 2026 | FLUX Non-Commercial |
| FLUX.2 [klein] Base 4B | 4 | 13 | 23.7 | Jan 2026 | Apache 2.0 |
| FLUX.2 [klein] Base 9B | 9 | 29 | 52.9 | Jan 2026 | FLUX Non-Commercial |
| GLM-Image | 16 | 23 · conditional | 35.8 | Jan 2026 | MIT |
| Qwen-Image-2512 | 20 | No publisher documents a memory path for this model | 57.7 | Dec 2025 | Apache 2.0 |
| LongCat Image | 6 | 17 · conditional | 29.3 | Dec 2025 | Apache 2.0 |
| Z-Image Turbo | 6 | 16 | 32.9 | Nov 2025 | Apache 2.0 |
| FLUX.2 [dev] | 32 | 24 · conditional | 177.6 | Nov 2025 | FLUX [dev] Non-Commercial |
| HunyuanImage 3.0 (Fal) | 80 | 240 | 168.7 | Sep 2025 | Tencent Hunyuan Community |
| HunyuanImage-2.1 | 17 | 24 · conditional | 173 | Sep 2025 | Tencent Hunyuan Community |
| SRPO | 11.9 | No publisher documents a memory path for this model | 47.6 | Sep 2025 | SRPO |
| Qwen-Image | 27.1 | 4 · conditional | 57.7 | Aug 2025 | Apache 2.0 |
24 of 91
37 more repositories with limited data
| FLUX.1-dev-FP8 | Publisher terms |
|---|---|
| FLUX.1-dev-NVFP4 | Publisher terms |
| FLUX.1-Kontext-dev-NVFP4 | Publisher terms |
| FLUX.2-dev-NVFP4 | Publisher terms |
| FLUX.2-klein-9b-fp8 | Publisher terms |
| FLUX.2-klein-9b-nvfp4 | Publisher terms |
| FLUX.2-klein-base-9b-fp8 | Publisher terms |
| FLUX.2-klein-base-9b-nvfp4 | Publisher terms |
| Hunyuan3D-2 | Publisher terms |
| Hunyuan3D-2.1 | Publisher terms |
| Hunyuan3D-2mv | Publisher terms |
| Hyper-SD | Not stated |
| InstantMesh | Apache 2.0 |
| LongCat-Image-Edit | Apache 2.0 |
| LongCat-Image-Edit-Turbo | Apache 2.0 |
| PhotoMaker-V2 | Apache 2.0 |
| Sana_1600M_1024px | Apache 2.0 |
| sd-turbo-amdnpu | Publisher terms |
| sdxl-base-amdnpu | OpenRAIL++ |
| sdxl-turbo_amdgpu | Publisher terms |
| sdxl-turbo-amdnpu | Publisher terms |
| sdxl-turbo-tensorrt | Publisher terms |
| shap-e | MIT |
| stable-cascade | Publisher terms |
No Download · GB column: repository sizes were not read for these models.
24 of 37