--- license: apache-2.0 base_model: black-forest-labs/FLUX.2-klein-base-4B tags: - flux - flux2 - lora - text-to-image - style widget: - text: "darkframe-slasher, an adult standing on a suburban porch at night" output: url: examples/porch.jpg - text: "darkframe-slasher, a boiler room full of pipes and steam" output: url: examples/boiler_room.jpg - text: "darkframe-slasher, an empty elm-lined street" output: url: examples/elm_street.jpg - text: "darkframe-slasher, a 1980s sedan in a driveway with the headlights on" output: url: examples/sedan.jpg - text: "darkframe-slasher, a bedroom window open to the fog" output: url: examples/bedroom.jpg - text: "darkframe-slasher, a motel corridor under a buzzing fluorescent" output: url: examples/motel.jpg --- # darkframe-slasher Style LoRA for a 1980s analog slasher look: film grain, deep red and metallic blue gels, heavy fog, crushed shadows, and a suburban night. | | | |---|---| | Base model | [`black-forest-labs/FLUX.2-klein-base-4B`](https://huggingface.co/black-forest-labs/FLUX.2-klein-base-4B) | | Trigger | `darkframe-slasher` | | Rank / alpha | 32 / 32 | | Training | ostris/ai-toolkit, 2500 steps, AdamW 8-bit, lr 1e-4 | | Weight keys | ComfyUI `diffusion_model.` prefix | Start the prompt with the trigger. The look is the fog, grain, and gel light. The rest of the prompt names the subject. ## Example images These are the step-2500 training samples. | Prompt | Image | |---|---| | `darkframe-slasher, an adult standing on a suburban porch at night` | ![porch](examples/porch.jpg) | | `darkframe-slasher, a boiler room full of pipes and steam` | ![boiler room](examples/boiler_room.jpg) | | `darkframe-slasher, an empty elm-lined street` | ![elm street](examples/elm_street.jpg) | | `darkframe-slasher, a 1980s sedan in a driveway with the headlights on` | ![sedan](examples/sedan.jpg) | | `darkframe-slasher, a bedroom window open to the fog` | ![bedroom](examples/bedroom.jpg) | | `darkframe-slasher, a motel corridor under a buzzing fluorescent` | ![motel](examples/motel.jpg) | ## Use on Fal The weights match FLUX.2 Klein 4B. Pass this repo, or the file URL, in `loras`. - Base, the model this LoRA was trained on: `fal-ai/flux-2/klein/4b/base/lora` (guidance 4, about 28 steps) - Distilled 4-step: `fal-ai/flux-2/klein/4b/lora` - Edit, base: `fal-ai/flux-2/klein/4b/base/edit/lora` - Edit, distilled: `fal-ai/flux-2/klein/4b/edit/lora` ```json { "prompt": "darkframe-slasher, an adult standing on a suburban porch at night", "guidance_scale": 4, "num_inference_steps": 28, "loras": [ { "path": "https://huggingface.co/deadvram/darkframe-slasher/resolve/main/darkframe-slasher.safetensors", "scale": 1.0 } ] } ``` Klein 9B and FLUX.2 dev use a different transformer, so this file will not load on those endpoints.