Workflow library
Explore published ComfyUI walkthroughs for LTX 2.5, MiniMax H3, and Krea 2. Each guide explains the workflow and links to the original video. Download resources stay with their source; no unverified files are mirrored here.
Showing 8 of 8 entries
Follow the Turbo LoRA workflow from a coarse first pass through learned latent upscaling and a second detail pass, with text-to-video and first/last-frame modes.
Track an object with SAM 3.1, clean its mask, and regenerate a cropped region before blending it back into the source video.
Arrange first and last frames, additional image guides, timed prompts, and video references in Director 2.0 to give a shot a clearer structure.
Set up the matching LTX 2.5 model stack in ComfyUI, choose dimensions and duration deliberately, and inspect the limits of the early results.
Assign clear jobs to reference images and video clips, then build a sequence using short continuation references instead of loading unnecessary frames.
Understand the FP8/INT8 model options, matching encoders and VAEs, native audio, and the resolution-duration tradeoff on the tutorial's 8 GB GPU.
Explore the identity-edit LoRA through restaging, object edits, and clothing references—with both the successful edits and the failure cases kept in view.
Explore image generation across environments, materials, portraits, and stylized scenes, then walk through the local FP8 setup and sizing controls.
Planning references, separate from the published video guides. These entries are not downloadable releases or claims of completed hardware testing.
Showing 4 of 4 entries
An editorial preview of a clean, memory-conscious node layout for learning how low-VRAM workflows are structured.
An editorial preview for documenting LTX 2.3 requirements, dependencies, and hardware constraints without making unverified performance claims.
A planned workflow for breaking image-to-video generation into understandable stages that can be tuned for limited VRAM.
A practical planning entry for matching model size, quantization, context length, and offloading choices to consumer hardware.