Civitai Model Matrix
The Civitai models supported by SoloDesk AI are the ones that have been derived from Stable Diffusion 1.x and Stable Diffusion XL.
With SD 1.5 and SDXL having already been tested on the reference machine (REF2026), it is likely that almost any SD1.x/SDXL model can be supported.
The models in the information matrix are a few that have been tested and known to work (or not work). These represent only a small subset of what can be found on the Civitai website. Therefore, it is by no means an exhaustive list. The model information matrix gets updated periodically as new models are added and new testing is performed.
| Model Name | Base Model | Disk (GB) | GPU Memory (GB) | Source | Input/Output | Supported | Notes |
|---|---|---|---|---|---|---|---|
| epiCRealism / Crystal Clear XL | Stable Diffusion XL | 6.8 | 8.2 | 277058?modelVersionId=1920523 | Text+Image+Mask-to-Image | Yes | Tested. |
| IntoRealism XL | Stable Diffusion XL | 6.8 | 8.2 | 1609320?modelVersionId=2650268 | Text+Image+Mask-to-Image | Yes | Tested. |
| Real Dream Classic SD1.5 | Stable Diffusion 1.5 | 2.1 | 3.2 | 462367?modelVersionId=717109 | Text+Image+Mask-to-Image | Yes | Tested. |
| AbsoluteReality | Stable Diffusion 1.5 | 2.1 | 3.2 | 81458/absolutereality | Text+Image+Mask-to-Image | Yes | Tested. |
| epiCRealism / Natural Sin | Stable Diffusion 1.5 | 2.1 | 3.2 | 25694?modelVersionId=160989 | Text+Image+Mask-to-Image | Yes | Tested. |
| HuslyoRealismXL | Stable Diffusion XL | 6.8 | 8.2 | 1802228/huslyorealismxl | Text+Image+Mask-to-Image | Yes | Tested. |
| GonzaLomo XL | Stable Diffusion XL | 6.8 | 8.2 | 1513492?modelVersionId=2439232 | Text+Image+Mask-to-Image | Yes | Only use Non-DMD version. |
| Dream Shaper XL | Stable Diffusion XL | 6.8 | 8.2 | 112902/dreamshaper-xl | Text+Image+Mask-to-Image | Yes | Tested. |
| Juggernaut XL | Stable Diffusion XL | 6.8 | 8.2 | 133005?modelVersionId=1759168 | Text+Image+Mask-to-Image | Yes | Tested. |
| Real Dream Pony 20 XL | Pony | 6.8 | 8.2 | 153568?modelVersionId=2129811 | Text+Image+Mask-to-Image | Yes | Tested. |
| Z Image Turbo | Z Image Turbo | 11.4 | 2168935?modelVersionId=2442439 | Text+Image+Mask-to-Image | No | Incomplete model. Missing Qwen3 text encoder. | |
| Jib Mix Illustrious Realistic | Illustrious-XL | 6.8 | 8.2 | 1255024/jib-mix-illustrious-realistic | Text+Image+Mask-to-Image | Yes | Tested. |
| Pikon-Realism | Stable Diffusion XL | 6.9 | 8.2 | 1855140/pikon-realism | Text+Image+Mask-to-Image | Uses non-standard scheduler. Advanced users can export to diffusers and modify scheduler config to LCM type. | |
| CyberRealistic | Stable Diffusion 1.5 | 2.1 | 3.2 | 15003/cyberrealistic | Text+Image+Mask-to-Image | Yes | Tested. |
| The Sixth Sense | Stable Diffusion XL | 6.9 | 8.4 | 1892984/the-sixth-sense-realistic-sdxl-bysilas | Text+Image+Mask-to-Image | Yes | Tested. |
To obtain Civitai models, the user downloads a large weight file (usually a safetensors file), and saves it to their own (well organized) archive. This is done directly with their web browser, not a downloader utility. Unlike models from Huggingface, models downloaded from Civitai are usually incomplete. For Stable Diffusion class models, the configuration data will be missing. As of SoloDesk 26.3, this missing configuration data can be "side loaded" from a local repository of known standard configurations. This means the user can directly use the models without any conversion, saving time and disk space. Some newer, larger models will be fragmented to the point of being unusable (missing not just the configuration, but also some of the model weights for components like text encoders and VAEs). Future innovations will be required to resolve these fragmentation issues.
Some models were trained on a non-standard configuration. For example, some models will use a different scheduler (or "sampler" as the Civitai community may call it). Advanced users can side load this model, export it to Diffusers format, and then edit the configuration files directly. Novice users might just avoid these models altogether.
Note that the GPU memory consumption will have some variance based on the mode in which the model operates. Different operational modes are text-to-image, image-to-image, image+mask-to-image, as well as bfloat16, float16 or float32.
View Supported Huggingface Models