🐙 Kraken WAN Prompt Splitter

WAN video is shot-based. You don't generate a "movie," you generate clips, and a good multi-shot video starts with a prompt that's been split into segments with each segment describing one shot. 🐙 Kraken WAN Prompt Splitter is the Kraken Tools node built for exactly that: type one multi-part idea, split it on a delimiter into up to five prompts, and add cinematic style blocks (shot type, lens, aperture, lighting, film stock, grade, composition, environment, texture) to each segment as you see fit.

It's part of Kraken Tools (krakenunbound/comfyui-kraken-tools), and it pairs with the pack's Kraken WAN Helper in the README's example graph: WAN Prompt Splitter → WAN Text Encode → WAN generation. The community rule for WAN is that prompting is instruction-style, not caption-style - a flat tag list underperforms a prose description of the shot. This node is a factory for well-formed shot prompts.

How it works

Two stages: split, then style.

The split stage takes your prompt and breaks it on delimiter - ---, ###, ***, ~~~, or ||| - into up to five segments (one per output). Separate your shots in the text with ---, and each becomes its own prompt. strip_whitespace (on) cleans the edges; keep_empty (off) drops empty segments so you don't get a blank Prompt 3. A persistent text (a character name, a brand, a recurring prop) can be added to every segment via combine_persistent and persistent_mode (append/prepend), so the subject stays consistent across shots.

The style stage is the cinematic control room. wan_pack gives you six pre-configured looks - Cinematic Natural, Moody Neon, Documentary Daylight, Epic Vista, Noir Classic, or None - that bundle a coherent set of the knobs below. apply_pack turns the pack on. If you'd rather dial it in yourself, every dimension is a dropdown:

• shot_type - extreme close-up, medium, cowboy, wide, POV, aerial, top-down…
• lens - 24mm through 200mm, plus anamorphic, tilt-shift, fisheye.
• aperture - f/1.4 dreamy bokeh through f/16 deep focus.
• lighting - golden hour rim, neon mix, studio 3-point, candlelight, moonlit blue hour…
• film - photoreal neutral, portra-like, cinestill-like, black-and-white, bleach-bypass…
• grade - warm amber teal, cool cyan steel, muted pastels, monochrome…
• composition / environment / texture / style_extra - framing rules, weather, grain, and a catch-all.

combine_style_presets (on) appends your chosen style block to every split segment; flip it off to get bare splits. separator sets how the pieces join.

Outputs

Five strings, named exactly as they'll be used:

• Prompt 1 … Prompt 5 - each a standalone WAN-ready prompt. Wire each into a WAN text encode (or into successive generations in your shot list). Unused ones will be empty strings if you wrote fewer than five segments.

The multi-output shape is the tell: this node is designed to be split out into five parallel WAN generations, or fed one-by-one into a sequential shot pipeline. Either way, the style stays consistent because the same persistent and style blocks went into every segment.

Install

Standard pack install. ComfyUI Manager → search "Kraken Tools" → Install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/krakenunbound/comfyui-kraken-tools
cd comfyui-kraken-tools
pip install -r requirements.txt

No models, no heavy deps - this node is string assembly.

Where people get burned

• The delimiter is your contract. --- is the default and it has to be on its own conceptual line between shots. If you use commas or periods to separate ideas, the node sees one giant segment and Prompt 1 gets everything. Use the delimiter deliberately.
• Packs and manual knobs can fight. If wan_pack is set to a pack and you also set a custom shot_type, the pack may override or append confusingly. Decide: pack, or manual - mostly. apply_pack / combine_style_presets are the switches for that.
• Persistent text on by default is not. combine_persistent defaults off, so a persistent string silently does nothing until you flip it. It reads like a bug the first time.
• Five outputs means five generations. If you only generate from Prompt 1 and wonder where shots 2–5 went - they didn't go anywhere; you need to wire the other outputs. That's the intended workflow, not a misfire.
• WAN wants prose, not tags. The default separator is ", " and the preset packs are prose phrases - good, that matches instruction-style WAN prompting. Don't feed the outputs through a tag-remover "to clean them up"; you'd be removing the part that works.

Multi-shot video is where WAN genuinely shines, and this node is the "type your idea once, get five directed shots" version of that. If your current WAN workflow is five separate prompt boxes with five different hand-typed styles, this replaces the box-and-forget with something coherent.