Pruna adapted the video model's text encoding, diffusion schedule and decoder for still images. The published checkpoint uses a short sampling schedule without classifier-free guidance, and an optional prompt upsampler expands briefs into richer conditioning.
Developers and artists can use the released Diffusers and ComfyUI integrations to compare prompting, generation and reference-based editing. The checkpoint's memory requirements vary with precision and output size, so the hosted playground offers a convenient first look.
