<ul><li>A fully ready Qwen3-4B-Thinking-2507 build. Compared to vanilla Qwen3-4B, it delivers noticeably better prompt adherence with Z-Image models and avoids common wording misinterpretations. Highly recommended for both inference and Z-Image LoRA training.</li><li>Path: ComfyUI_windows_portable\ComfyUI\models\text_encoders\</li><li>Qwen3-4B-Thinking-2507 USAGE:<pre><code>[ANCHOR: who / what exists]
[ROLE or STATE: what defines them conceptually]
[ACTION or POSTURE: what they are doing or how they are positioned]
[RELATIONSHIP: how they relate to space, objects, or viewer]
[ENVIRONMENT: where this takes place, minimally]
[INTENT: what the image is meant to communicate]
[LIGHTING: chosen to support the intent]
[CAMERA / FRAMING: how the viewer perceives it]
[STYLE RESTRAINTS: what it should resemble, softly]
[CONSTRAINTS: what must be avoided]

example:
a single adult man,
calm and self-contained rather than expressive,
standing upright with relaxed posture,
positioned slightly off-center to create quiet tension,
inside a simple, uncluttered interior space,
the focus is on presence and character rather than action,
soft indirect light so that facial features remain natural,
eye-level camera, medium framing from the chest up,
realistic but understated photographic style,
no exaggerated emotion, no stylization, no dramatic effects

example 2:
[SUBJECT / ANCHOR], 
[TRAIT / MOOD / PERSONALITY], 
[ACTION / POSTURE / STATE], 
[POSITION / RELATION TO SPACE / COMPOSITION], 
[ENVIRONMENT / SETTING], 
[INTENT / WHAT THE IMAGE SHOULD CONVEY], 
[LIGHTING / ATMOSPHERE], 
[CAMERA / FRAMING / PERSPECTIVE], 
[STYLE / ARTISTIC DIRECTION], 
[FORM CLARITY / SHAPE / TEXTURE / COLOR DIRECTIONS]

example:
a single adult man, 
calm and self-contained, 
standing upright with relaxed posture, 
positioned slightly off-center to create quiet tension, 
inside a simple, uncluttered interior space, 
showing presence and character through posture and expression, 
soft indirect light to enhance facial features naturally, 
eye-level camera, medium framing from the chest up, 
photographic style with subtle tones and understated textures, 
featuring clear forms, natural proportions, and readable visual composition</code></pre></li></ul><ul><li>You use this inside of your positive prompt; meaning the example part only. the explaining part is just for you to understand the layout not the text encoder</li></ul>**Please note that Qwen3-4B-Thinking-2507 is just experimental with this model but with right tweaks it can provide great outputs and any trained lora on the vanilla qwen3_4b will not function properly under this encoder so you will need to retrain using this text encoder.<ul><li>full training thread can be found <a target="_blank" rel="ugc" href="https://civarchive.com/articles/24403/my-little-research-about-z-image-lora-training-fp32-model-different-text-encoders-upscaling">here</a></li><li>training config for AiToolKit with Qwen3-4B-Thinking-2507 text encoder Can be found <a target="_blank" rel="ugc" href="https://pastebin.com/DAg870DU">here</a>.</li></ul>

Qwen3-4B-Thinking-2507 text encoder

a merge of Qwen3-4B-Thinking-2507

qwen34BThinking2507_v10.safetensors

v1.0

merged Qwen3_4B_Thinking_2507 text encoder <a target="_blank" rel="ugc" href="https://pastebin.com/jKP0gaMd">workflow</a>huggingface with full repo files : <a target="_blank" rel="ugc" href="https://huggingface.co/Capitan01R/qwen-thinking-merged/tree/main/qwen_thinking_merged">files to load via llm loader</a>recommended settings <a target="_blank" rel="ugc" href="https://www.reddit.com/r/StableDiffusion/comments/1q70hi6/better_outputs_overall_and_understanding/">here</a>

Qwen3_4B_Thinking_2507

<ul><li>A fully ready Qwen3-4B-Thinking-2507 build. Compared to vanilla Qwen3-4B, it delivers noticeably better prompt adherence with Z-Image models and avoids common wording misinterpretations. Highly recommended for both inference and Z-Image LoRA training.</li><li>Path: ComfyUI_windows_portable\ComfyUI\models\text_encoders\</li><li>Qwen3-4B-Thinking-2507 USAGE:<pre><code>[ANCHOR: who / what exists]
[ROLE or STATE: what defines them conceptually]
[ACTION or POSTURE: what they are doing or how they are positioned]
[RELATIONSHIP: how they relate to space, objects, or viewer]
[ENVIRONMENT: where this takes place, minimally]
[INTENT: what the image is meant to communicate]
[LIGHTING: chosen to support the intent]
[CAMERA / FRAMING: how the viewer perceives it]
[STYLE RESTRAINTS: what it should resemble, softly]
[CONSTRAINTS: what must be avoided]

example:
a single adult man,
calm and self-contained rather than expressive,
standing upright with relaxed posture,
positioned slightly off-center to create quiet tension,
inside a simple, uncluttered interior space,
the focus is on presence and character rather than action,
soft indirect light so that facial features remain natural,
eye-level camera, medium framing from the chest up,
realistic but understated photographic style,
no exaggerated emotion, no stylization, no dramatic effects

example 2:
[SUBJECT / ANCHOR], 
[TRAIT / MOOD / PERSONALITY], 
[ACTION / POSTURE / STATE], 
[POSITION / RELATION TO SPACE / COMPOSITION], 
[ENVIRONMENT / SETTING], 
[INTENT / WHAT THE IMAGE SHOULD CONVEY], 
[LIGHTING / ATMOSPHERE], 
[CAMERA / FRAMING / PERSPECTIVE], 
[STYLE / ARTISTIC DIRECTION], 
[FORM CLARITY / SHAPE / TEXTURE / COLOR DIRECTIONS]

example:
a single adult man, 
calm and self-contained, 
standing upright with relaxed posture, 
positioned slightly off-center to create quiet tension, 
inside a simple, uncluttered interior space, 
showing presence and character through posture and expression, 
soft indirect light to enhance facial features naturally, 
eye-level camera, medium framing from the chest up, 
photographic style with subtle tones and understated textures, 
featuring clear forms, natural proportions, and readable visual composition</code></pre></li></ul><ul><li>You use this inside of your positive prompt; meaning the example part only. the explaining part is just for you to understand the layout not the text encoder</li></ul>**Please note that Qwen3-4B-Thinking-2507 is just experimental with this model but with right tweaks it can provide great outputs and any trained lora on the vanilla qwen3_4b will not function properly under this encoder so you will need to retrain using this text encoder.<ul><li>full training thread can be found <a target="_blank" rel="ugc" href="https://civitai.com/articles/24403/my-little-research-about-z-image-lora-training-fp32-model-different-text-encoders-upscaling">here</a></li><li>training config for AiToolKit with Qwen3-4B-Thinking-2507 text encoder Can be found <a target="_blank" rel="ugc" href="https://pastebin.com/DAg870DU">here</a>.</li></ul>

Files with SHA256: c64f43de489bf2421f8a544951c405c7fbb37e518c382bb741444aff54eff9c7