Adjusted this workflow for a clean single-pass run. Do a second pass only if you like wasting compute.

Key Features

• Supports up to 5 reference images.

• Audio references for up to 2 speakers.

• Preserves multiple characters, clothing, objects, and backgrounds natively.

• Slot embeddings and native self-attention retrieval.

• Multi-subject and subject-object composition tailored for LTX-2.5.

Prompt Rules (Don't mess this up)

• Describe each reference image clearly in the prompt.

• Use consistent labels (Image 1, Image 2, etc.).

• Define actions and spatial relationships explicitly.

• Specify which reference feeds the character, object, clothing, or background.

• Match audio references to the correct subject image.

Links & Credits

• Character Layouts Workflow: Civitai

• MSR LoRA: Hugging Face (Credits to LiconStudio for the solid work on the LoRA).

• Base models belong to their respective creators; this is a custom workflow build.

• Compatible models support NSFW generation.

• Message directly if you run into configuration issues.

If this workflow saves you time: Buy me a coffee