Anima Artist Mix Conditioning

Anima is an artist-tag machine - the whole reason people switch is that naming an artist actually steers the output. But what if you want two artists at once? Naively stacking artist_a, artist_b usually makes the model blend them into mush or pick one and abandon the other. Anima Artist Mix Conditioning is the node that does artist blending properly: it takes a normal prompt plus separate artist tags and builds a positive CONDITIONING with real mix modes. And it's standalone - no Prompt Studio, no dict socket required.

What it does

You feed it a CLIP, a base prompt, and artist_tags. It positions the artists according to ANIMA ordering rules (correct) or pins them front/back, then encodes the result with one of the mix modes. The single output is positive CONDITIONING, ready for a KSampler.

The artist-tag syntax gives you two controls:

• (artist:1.2) - a per-artist mix weight.
• [[artist_a, artist_b:0.7]] - keep multiple artists in one mix branch; the final :0.7 is a conditioning mix weight, not a prompt-string weight.

The mix modes (and their real cost)

Every mode's tooltip tells you its branch cost - worth reading, because each positive branch is extra conditioning work per step:

• prompt (default) and average - one branch, fastest, stable. Artist text stays inline / gets averaged as conditioning. Start here.
• delta_rms - one branch, mixes artist deltas from the base prompt and restores RMS style energy. Usually stronger than average.
• hybrid - keeps your strongest top_k artists as exact branches and compresses the tail with delta_rms. The recommended balance for several artists.
• clustered - groups similar artist deltas into cluster_count compressed branches; for many artists.
• exact and composite_exact - N (or N+1) branches, the most faithful per-artist output, and the most expensive.
• late_exact, average_late_exact, scheduled_average - change mix weights across timestep ranges instead of one flat mix.

The README's advice is the sane path: start with artist_position=correct and mode prompt or average, reach for hybrid when several artists need clearer separation, and use exact only when the artist count is small or per-artist separation matters more than conditioning cost. Watch the branch count before you raise tuning strength.

The tuning knobs - artist_mix_style_gain (1.35 default), artist_mix_rms_scale_cap (2), artist_mix_exact_top_k (4), artist_mix_dominant_threshold (0.25) - all have tooltips explaining exactly which branch or weight they affect. Defaults are good; adjust the ones whose tooltip names your problem.

When to use this vs the dict flow

If your workflow uses Prompt Studio Advanced v2, artist mixing is built into the prompt data and Prompt Data Conditioning handles it there. Use this standalone node when you don't use Prompt Data, or when you want to test artist-tag conditioning in isolation without dragging the editor in.

Install and gotchas

Pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/n0va39/ComfyUI-EasyUseAnima
cd ComfyUI-EasyUseAnima && pip install -r requirements.txt

Restart ComfyUI; Manager: "ComfyUI EasyUse Anima".

The trap here is performance, not correctness: exact and composite_exact look seductive and will noticeably slow generation on midrange cards - Anima is already slow, and N conditioning branches make it slower. If a mix looks like one artist ate the other, that's usually a weight problem: raise the (artist:weight) on the one getting eaten before you switch modes. And remember the group syntax is [[a, b:weight]], not a nested weight - (a:0.35) is how you weight one artist inside a group.