ArtificialGuyBR

Home / Blog / Training LoRAs

Cinematic Filmstill LoRA for Z-Image Turbo: Film Still Style Generation

Generate cinematic film still images with Z-Image Turbo using this LoRA by ArtificialGuyBR. Trigger words, settings, and usage guide.

1 sources cited Training LoRAs

On this page

  1. What This Model Does
  2. Trigger Words
  3. Recommended Settings
  4. Example Workflow (ComfyUI)
  5. License
  6. Common Errors and Fixes
  7. FAQ
  8. Sources

Cinematic Filmstill LoRA for Z-Image Turbo

What This Model Does

This LoRA (Low-Rank Adaptation) fine-tunes the Z-Image Turbo text-to-image model to produce images with a distinct cinematic film still aesthetic. The training dataset consisted of film still cinematic style images, resulting in outputs that mimic the look of movie frames — dramatic lighting, composition, and color grading typical of cinema.

The model was trained with GPU compute sponsored by Redmond.AI, and the author (ArtificialGuyBR) publishes it under the permissive Apache 2.0 license.

Trigger Words

Include these in your prompt to activate the film still style:

Cinematic. Film Still.

Or use the uppercase variants:

CINEMATIC FILM STILL

The model card shows both formats work; the instance prompt used during training was CINEMATIC, FILM STILL.

Typical Z-Image Turbo settings apply: low step counts (4–8 steps), CFG 1.0–2.0, and the model's native resolution.

Example Workflow (ComfyUI)

  1. Load Z-Image Turbo as your base checkpoint
  2. Add this LoRA via a LoraLoader node
  3. Set LoRA strength to 0.8–1.0
  4. Include trigger words in your positive prompt
  5. Generate with 4–8 steps, CFG 1.0–2.0

License

Apache License 2.0 — permissive for commercial and non-commercial use. See the HF model card for the full license text.

Common Errors and Fixes

Error | Cause | Fix | Source

|-------|-------|-----|--------

Style not appearing | Missing trigger words | Add Cinematic. Film Still. to prompt | HF model card Wrong base model | Using SDXL/SD1.5 instead of Z-Image Turbo | Load Tongyi-MAI/Z-Image-Turbo as checkpoint | HF model card Poor quality at high LoRA weight | Over-saturation | Reduce LoRA weight to 0.6–0.8 | General LoRA practice

FAQ

Q: Does this work with Stable Diffusion XL or 1.5? A: No. This LoRA is trained specifically for Tongyi-MAI/Z-Image-Turbo and must be used with that base model.

Q: What are the recommended generation parameters? A: Z-Image Turbo is a fast model — use 4–8 steps, CFG 1.0–2.0, and the trigger words. ComfyUI is recommended.

Q: Can I use this commercially? A: Yes, Apache 2.0 permits commercial use with attribution.

Q: Where can I download the weights? A: From the Files & versions tab on HuggingFace in Safetensors format.

Sources