CivArchive
    Capitan Conditioning Enhancer – Better Prompt Adherence for Z-Image Workflows/Qwen-edit-2511 - v1.0.2
    Preview 117549084
    Preview 117549108
    Preview 117549100

    UPDATE JAN 13TH, 2026

    Quick update on my Capitan Conditioner Pack, original post here if you missed it.

    The basic Conditioning Enhancer is unchanged (just added optional seed for reproducibility).

    New addition: Capitan Advanced Enhancer – experimental upgrade for pushing literal detail retention harder.

    It keeps the same core (norm → MLP → blend → optional attention) but adds:

    • detail_boost (sharpens high-frequency details like textures/edges)

    • preserve_original (anchors to raw embeddings for stability at high mult)

    • attention_strength (tunable mixing – low/off for max crispness)

    • high_pass_filter (extra edge emphasis)

    Safety features like clamping + residual scaling let you crank mlp_hidden_mult to 50–100 without artifacts.

    Best use: Stack after basic, basic glues/stabilizes, advanced sharpens literally.
    Start super low strength (0.03–0.10) on advanced to avoid noise.

    Repo : https://github.com/capitan01R/Capitan-ConditioningEnhancer
    Install via Comfyui Manager or git clone.

    Also qwen_2.5_vl_7b supported node is released. (usually used for Qwen-edit-2511), you can just extract to your custom nodes: latest release

    Full detailed guide is available in the repo!!

    Full examples and Grid examples are available for both basic and advanced nodes in the repo files basic & advanced

    Let me know how it performs for you!

    Thanks for the feedback on the first version, appreciate it!!


    Just released Capitan Conditioning Enhancer – a lightweight ComfyUI custom node that improves the 2560-dim embeddings coming from Qwen3-4B (and now also Qwen 2.5 VL 7B).

    It sits between text encoder and KSampler and helps with: • better coherence & scene unity • stronger prompt adherence • more consistent mood & detail retention

    What it does (in brief):

    • Per-token normalization (usually keep on)

    • 2-layer MLP refiner (starts very subtle)

    • Optional 8-head self-attention (helps distant prompt parts connect)

    Quick-start settings Everyday / stable look Strength: 0.00 – 0.10 Normalize: True Self-Attn: True MLP hidden mult: 2–4

    Detail / literal stack (my favorite)

    1. Glue node: Strength 0.05, Self-Attn True, Mult 2–4

    2. Detail node: Strength -0.10, Self-Attn False, Mult 40–50

    Installation git clone https://github.com/capitan01R/Capitan-ConditioningEnhancer.git into ComfyUI/custom_nodes/ → restart ComfyUI

    OR

    • install it via Comfyui manager lookup: "Capitan-ConditioningEnhancer"

    Qwen 2.5 VL 7B support → grab the release here: here

    Examples (grids + drag-and-drop workflows) added to repo: here

    Reddit thread: here

    Let me know what settings work best for you or if you run into any issues!

    Happy prompting!

    GitHub: https://github.com/capitan01R/Capitan-ConditioningEnhancer

    Description

    bug fixes added smarter memory management

    Workflows
    Other

    Details

    Downloads
    72
    Platform
    CivitAI
    Platform Status
    Deleted
    Created
    1/15/2026
    Updated
    2/8/2026
    Deleted
    2/8/2026

    Files

    capitanConditioningEnhancer_v102.zip

    Mirrors