Anima Prompt Data Conditioning

The AiO Generator does conditioning for you, invisibly, inside its one-node flow. Anima Prompt Data Conditioning is the same logic exposed as a normal node, for when you want to steer the pipeline yourself - a custom sampler, a latent you made, or a workflow that isn't using the all-in-one generator at all.

What it does

It reads an EASYUSE_ANIMA_PROMPT_DATA dict by key and produces everything the sampling stage needs:

• positive and negative CONDITIONING, encoded with the CLIP you give it
• latent_image - an empty latent built from the prompt-data width and height, batch size fixed to 1
• model - passed through, or patched with Anima Mod Guidance when that's enabled

So a single node replaces the text-encode + empty-latent trio you'd otherwise wire by hand, and it does it from the same dict the editor already carries.

The two features that make it interesting

Anima Mod Guidance. This is the model patch from ComfyUI-Spectrum-KSampler that pushes quality tags through a separate guidance path instead of bloating the positive prompt. mod_guidance_mode decides whether it's used: prompt_data follows the flag stored in the dict, enabled forces it on, disabled bypasses the patch. mod_w_profile picks the per-block profile - off, or one of step_i8_skip27, step_i14, uniform_w3. If Spectrum isn't installed and a profile other than off is active, that's where you'll get the failure.

Artist mix. The artist-field text lives in the dict as structured artist data. With artist_mix_mode at prompt_data it follows the dict; otherwise you can override. Modes range from prompt and average (one positive branch, fast, stable) up through hybrid, clustered, exact, late_exact and scheduled_average, with the branch cost spelled out right in the tooltips - exact costs N positive branches and clustered costs about cluster-count plus dominant artists. The defaults are sane: artist_mix_start_percent 0.5, style_gain 1.35, rms_scale_cap 2, exact_top_k 4.

The inputs and outputs that matter

Required: model, clip, and EASYUSE_ANIMA_PROMPT_DATA. The rest are the mod-guidance and artist-mix controls above. For a first workflow, leave mod_guidance_mode on prompt_data, mod_w_profile on step_i8_skip27, and artist mix on prompt_data - that's the pack's tuned default path.

Outputs: model (patched or passed through), positive, negative, latent_image. Wire them to a KSampler and you have the AiO generator's first-pass logic minus the training wheels.

Install and gotchas

Pack install, same as always:

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

Restart ComfyUI; Manager finds it under "ComfyUI EasyUse Anima".

The main trap is the dependency: Mod Guidance only exists if ComfyUI-Spectrum-KSampler is installed, and the README is explicit that the latest version is recommended. Second trap: the latent output is always batch 1 - if you were hoping to feed it a batch you made yourself, that's not this node's job; generate your own LATENT and only use this node for conditioning. And if artist-mix mode differs from what you set in the editor, remember the dict is the source of truth when mode is prompt_data - change it in the editor, not here.