IAMCCS WAN FLF Pair From Timeline

The whole trick of IAMCCS's shotboard pipeline is that every chunk of a long video is an FLF - a "first-last-frame" transition. You generate from image A, and you tell the sampler it should arrive at image B. IAMCCS_WanFLFPairFromTimeline ("IAMCCS WAN FLF Pair From Timeline") is the node that picks which A→B pair you're working on for a given chunk, pulls the actual images off disk, and hands you everything else that chunk needs to render.

You feed it the timeline_plan from IAMCCS_WanCineInfoPure plus a chunk_index, and out come the two images, the chunk's prompt, its length, its motion setting, and the frame rate. It's the per-iteration heart of the shotboard loop, and the natural place to start when you want to understand how this pack builds a long video: one planner decides the shots, this node serves them up one at a time.

How it works

The node looks up flf_pairs[chunk_index] in the plan, resolves the start and end image paths, loads them with the board's resize settings (crop by default, 16-pixel multiple), and reads the chunk's prompt, length (falling back to the board's total duration), and motion (clamped to the 1.0–2.0 range). It also pulls the prompt-relay data for that chunk: relay_local_prompts, relay_segment_lengths, relay_max_frames, and relay_epsilon - the bits a prompt-relay system needs to splice per-chunk text into the generation.

The chunk_index input is marked lazy, so if it's wired to a loop's iteration counter the node only re-runs when the index actually changes - which is what you want in a 40-chunk render.

The outputs that matter

• start_image / end_image - feed these to your anchor and end-frame encoder. The A→B contract your sampler has to honor.
• prompt - the per-chunk text, ready for a text encoder.
• frames, motion, frame_rate - pacing and motion amplitude for this chunk, wired into the motion/conditioning nodes.
• chunk_index, chunk_count - pass-throughs so downstream nodes know where they are in the sequence.
• The relay_* outputs - only relevant if your pipeline uses prompt relay; otherwise ignore them.

Install and practical notes

Standard IAMCCS-nodes install - ComfyUI Manager (search "IAMCCS") or:

cd ComfyUI/custom_nodes
git clone https://github.com/IAMCCS/IAMCCS-nodes.git

Restart, done. No model files, no pip extras. ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8.

Two things trip people up. First, the start/end images are loaded from the paths baked into the shotboard, so if you move files after planning, the node throws on a missing path - check the report output for the resolved path. Second, prompt falls back to the board's global prompt when a chunk has none, so an "empty" chunk quietly renders with the default text. If a segment looks wrong, confirm the pair you're on: chunk_index is 0-based, and mismatched loop counters are the classic off-by-one.