Krea Moodboard Apply

If you use this pack at all, you end up here. Krea Moodboard Apply is the node that takes your normal image prompt, bolts a moodboard's style text onto it, and hands the combined result to Krea 2. Everything else in the pack - Search, Random, Style, the browsers, Mashup - exists to feed this one node. Think of it as the funnel.

Why it exists is worth knowing. Krea's hosted product runs a style-reference and moodboard module that was explicitly not open-sourced when Krea 2's weights shipped in June 2026 - the company said as much in its AMA. So the community's workaround was to copy styles out of the public moodboard gallery by hand and ask an LLM to merge them into their prompt. That works, and it's exactly the chore this pack automates: 3,549 moodboards, rewritten into clean prompt style blocks, applied for you.

How it works

It's plain string concatenation, which is the whole trick of the pack. Nothing here loads Krea 2, patches the model, or touches conditioning tensors - the node takes the positive and negative STRING outputs from another Krea Moodboard node and merges them with your own text:

• your prompt plus the moodboard_positive block, joined with a newline (or comma if you set separator to comma)
• your negative_prompt plus the moodboard_negative guidance, comma-joined
• metadata_json just passes through untouched, so you can chain the same board into a Mashup later

The moodboard text itself already carries a style-only guardrail from the source node - an instruction telling the model to apply the board only as visual treatment and never to add or change your subject. That guardrail is the difference between "render my cat in this style" and "render a lone figure because the moodboard mentioned one."

The inputs that matter

• prompt - your main image prompt. This is where you describe the actual subject.
• moodboard_positive / moodboard_negative - wire these from the positive and negative outputs of Search, Random, Style, or a browser node.
• style_only - flips the node to output only the moodboard style text, no user prompt. Handy if you're feeding the style block into a different prompt-enhancement setup instead of the Krea 2 path.
• separator - newline or comma, how the style block gets glued on.

One honest gotcha: the strength input here is a label. The author's own tooltip says it's "reserved for workflow clarity" - the actual strength is applied by the source node that produced the moodboard text. If you want a stronger or more concise style block, set strength on the Search/Random/Style node upstream, not here.

Where it wires in

The outputs are positive, negative, and metadata_json. In the official Krea 2 Turbo workflow, connect positive to the Krea-2 Turbo subgraph's Text String (User Prompt) input, before any prompt enhancement and before the Qwen3-VL Krea2 text encode. If your workflow exposes a negative prompt or CFG path, connect negative there; if it doesn't, leave it unused - the model runs at CFG 1 anyway and the negative path is optional.

Install

It ships in the Krea Moodboards pack. Easiest route is ComfyUI Manager - search "Krea Moodboards" (or comfyui-krea-moodboards) and install. Manual install is the usual:

cd ComfyUI/custom_nodes
git clone https://github.com/Andro-Meta/ComfyUI-Krea-Moodboards.git

Restart ComfyUI. No Python dependencies beyond the standard library, no model downloads, no API key - this pack is prompt text only, so it can't brick your setup or fight your VRAM. The nodes appear under Andro.Meta / Moodboards.

Common issues

The most common mistake is setting strength here and wondering why nothing changed - set it upstream. Second most common: expecting the node to do something with an image. It doesn't; it's text in, text out, and that's the point. If your output looks like the moodboard replaced your subject rather than styled it, you're probably not feeding the guardrailed positive output from a real source node - don't hand-type moodboard text into the box, let the source node generate it.