Anima Artist Impact Map (A/B Diff)

Some artists change your render, and some change almost nothing - and you can't always tell by looking at one image. AnimaArtistImpactMap is the A/B diff tool for the Anima mixer. Give it two same-seed renders and it produces a visual comparison plus a numeric verdict, so "did that artist/setting do anything?" becomes a number and a heatmap instead of a gut feeling.

What you give it

• image_a - the reference render (e.g. mixer off, or the no-mixer baseline from AnimaArtistABVariants).
• image_b - the test render (e.g. mixer on, or a solo artist), same seed and same size. Same-seed/same-size is the whole point; a different seed diff would be noise.
• layout - triptych (default): [A | B | change overlay]. overlay: grayscale B with the change heatmap on top. heatmap: heatmap only.
• auto_gain (default on) - scales the heatmap so the strongest change is fully visible. Turn it off (and set gain yourself) when you want to compare heatmaps across different runs on one consistent scale.

What you get back

• visualization - the IMAGE, in the chosen layout.
• report - a plain-language STRING: impact score, changed-area %, composition-vs-texture and luminance splits, and a verdict. When an artist or setting changed nothing, it says so directly ("no visible change") instead of leaving you to interpret a 0.3% diff.
• impact_score - a FLOAT, so you can even route it into other logic if you're building something fancier.

Install

Manager → search "Anima Artist Mixer Forge", or:

cd ComfyUI/custom_nodes
git clone https://github.com/peter119lee/Anima-Artist-Mixer-Forge

then restart. No extra dependencies. Remove the original Anima-Artist-Mixer if present (duplicate node names shadow each other).

My take

The composition-vs-texture split is the part I'd actually trust - it separates "the artist changed the whole layout" from "the artist just recolored things," which are very different problems to fix. And the plain-language verdict matters more than it sounds: with ABVariants producing up to a dozen renders, you want the tool to tell you which diffs are real, not a number you have to interpret against your own baseline. The honest limit: it's a pixel diff, not an aesthetic judgement. A small pixel change can be a massive style change if it lands in the right place - use the heatmap to see where the change is, not just how big.