Hugging Face's logo

BennyDaBall
/
Z-Image-Engineer-V6

Text Generation
Transformers
Safetensors
English
qwen3
prompt-engineering
image-generation
z-image
z-image-turbo
text-encoder
comfyui
lm-studio
conversational
text-generation-inference
Model card Files
xet
 Community
BennyDaBall commited on
Commit
496b168
·
verified ·
1 Parent(s): 9aee4d8

Add files using upload-large-folder tool

Browse files
Files changed (3) hide show
  1. HASHES.sha256 +1 -8
  2. README.md +17 -23
  3. RELEASE_MANIFEST.json +6 -102
HASHES.sha256 CHANGED
@@ -6,14 +6,7 @@ D8AD1C43FD8C76F1EBDAD11C85D494A474F9CFE9F83AF5F72F590AE3852315A2 evidence/galle
6
  C4D12692AE5CEFA9B7E61C2A581062F6B4B06183165EB2BCEEE9E11F26B82308 model-00001-of-00003.safetensors
7
  93A4CAF2F35B815178DB5CE43C9FD5E06E3EF836F5CBEEE7C690961D95DA653B model-00002-of-00003.safetensors
8
  4B3EF3D52BCAD649213FD2035D94DF48CBF2FD670250EC1FD8E1748072ECDBF2 model-00003-of-00003.safetensors
9
- BF31DA5A1F64F1D7F9AF7C692C82296ED9B2AD59076588BF297E446D6FF54C1C README.md
10
  BE75606093DB2094D7CD20F3C2F385C212750648BD6EA4FB2BF507A6A4C55506 tokenizer.json
11
  154E5FF1E7C152D964EDF30DA854EA62465C767719AC8E97E58BABF2D4FA9079 tokenizer_config.json
12
  34126E2486E389F28C11693C2E51641199FB5B53E3E7D6BFA75A6E967C11D3CF V6_SYSTEM_PROMPT.md
13
- 20DAB6305B76B28808FAD740C7107878DEEC63688E1B318F7BB3A7F707220B0D Z-Image-Engineer-V6-F16.gguf
14
- A39695B6714FC4A0A86965F5B2FB8B0CBEF774165EEC8FB9B2379FBEDD86838A Z-Image-Engineer-V6-MXFP4.gguf
15
- E3F493D971677BA181F67C888AD41E25FD34448BF7EEA03A84F4114EE021B9E3 Z-Image-Engineer-V6-Q3_K_M.gguf
16
- D666E619EDB2D6DCF2DF013540B22E2592C4FBADB9007B3FB89D4BBE0C4C7C67 Z-Image-Engineer-V6-Q4_K_M.gguf
17
- 0FAB79F032AA34BAAC8607FF8BA720DFB95A0D9A44026DE79288F3FD25A66A05 Z-Image-Engineer-V6-Q5_K_M.gguf
18
- A27D6723816462EA1368093A76E9013E996BD4B731EF87327334D50D6DD9534C Z-Image-Engineer-V6-Q6_K.gguf
19
- DC4F5476A0F804A7DB73EDA164C0503CDA93858F3EABDE9EA36C68EEDCBA306C Z-Image-Engineer-V6-Q8_0.gguf
 
6
  C4D12692AE5CEFA9B7E61C2A581062F6B4B06183165EB2BCEEE9E11F26B82308 model-00001-of-00003.safetensors
7
  93A4CAF2F35B815178DB5CE43C9FD5E06E3EF836F5CBEEE7C690961D95DA653B model-00002-of-00003.safetensors
8
  4B3EF3D52BCAD649213FD2035D94DF48CBF2FD670250EC1FD8E1748072ECDBF2 model-00003-of-00003.safetensors
9
+ AF0D74388A53EF4A9B37CF98B05922F1D5D3A888C1C6BBB95D883D89AC423760 README.md
10
  BE75606093DB2094D7CD20F3C2F385C212750648BD6EA4FB2BF507A6A4C55506 tokenizer.json
11
  154E5FF1E7C152D964EDF30DA854EA62465C767719AC8E97E58BABF2D4FA9079 tokenizer_config.json
12
  34126E2486E389F28C11693C2E51641199FB5B53E3E7D6BFA75A6E967C11D3CF V6_SYSTEM_PROMPT.md
 
 
 
 
 
 
 
README.md CHANGED
@@ -13,7 +13,6 @@ tags:
13
  - z-image
14
  - z-image-turbo
15
  - qwen3
16
- - gguf
17
  - text-encoder
18
  - comfyui
19
  - lm-studio
@@ -31,7 +30,7 @@ tags:
31
  | **Base Model** | `Tongyi-MAI/Z-Image-Turbo` |
32
  | **Library** | `transformers` |
33
  | **Pipeline Tag** | `text-generation` |
34
- | **Format** | GGUF, HF Safetensors |
35
 
36
  ---
37
 
@@ -39,7 +38,7 @@ The **Z-Engineer** returns, fully rebuilt around the **SMART DoRA** training sys
39
 
40
  Yes, we jump from V4 to V6. Unlike the usual guy math, this one actually brought the extra two inches.
41
 
42
- **Z-Image-Engineer V6** is a fine-tuned 4B Qwen text encoder (`Tongyi-MAI/Z-Image-Turbo`) optimized for dual-role performance: a local prompt-enhancement model for LM Studio, and a direct drop-in replacement text encoder for ComfyUI.
43
 
44
  ![Z-Image-Engineer V6 simple A/B with rewrites](evidence/gallery_z_image_engineer_v6_simple_ab_with_rewrites_CONTACT.png)
45
 
@@ -49,12 +48,12 @@ Yes, we jump from V4 to V6. Unlike the usual guy math, this one actually brought
49
 
50
  V6 transforms minimal seed prompts into rich, highly structured visual narratives. It adds explicit scene composition, lighting direction, material texture, and depth separation while stripping out empty prompt sludge like *"8k, masterpiece, trending on ArtStation."*
51
 
52
- It can also be used directly as a Z-Image text encoder. Drop the GGUF into ComfyUI, load it with `CLIPLoaderGGUF`, set the type to `lumina2`, and compare it against the stock `qwen_3_4b.safetensors`.
53
 
54
  ### Key Use Cases
55
 
56
  - **Prompt Enhancement:** Upgrade simple concepts into descriptive, high-fidelity visual prompts locally.
57
- - **Text Encoder Swap:** Replace the stock Z-Image Qwen text encoder in ComfyUI to generate different conditioning from the same seed.
58
  - **Hybrid Mode:** Use V6 to rewrite your prompt, then use V6 again to encode it. It writes the scene and drives the image model.
59
  - **Private Local Workflow:** Built for LM Studio, ComfyUI, and `llama.cpp`. No API logs, no external telemetry.
60
 
@@ -89,7 +88,7 @@ V6 was not a simple one-and-done training run. The final architecture is a blend
89
 
90
  ### LM Studio: Prompt Enhancement
91
 
92
- Download your preferred GGUF quant, load the model, and prompt it directly. No complex system prompt is required.
93
 
94
  ```text
95
  Enhance this image prompt for Z-Image Turbo: a unicorn
@@ -99,10 +98,11 @@ The comparison examples were generated from direct LM Studio user requests like
99
 
100
  ### ComfyUI: Direct Encoder Swap
101
 
102
- 1. Place the GGUF file into `ComfyUI/models/text_encoders/`.
103
- 2. Add a `CLIPLoaderGGUF` node.
104
- 3. Set model type to `lumina2`.
105
- 4. Use it where the stock Z-Image Qwen text encoder would normally go.
 
106
 
107
  Optional workflow repo:
108
 
@@ -115,7 +115,7 @@ The raw GGUF works without the node.
115
  ```text
116
  UNET: z_image_turbo_bf16.safetensors
117
  VAE: ae.safetensors
118
- Text Encoder: Z-Image-Engineer-V6-Q8_0.gguf
119
  Resolution: 1024x1024
120
  Steps: 8
121
  CFG: 1.0
@@ -136,23 +136,17 @@ Shift: 3.0
136
  | **Rank / Alpha / Dropout** | 64 / 64 / 0.03 |
137
  | **Target Modules** | `q_proj`, `k_proj`, `v_proj`, `o_proj`, `gate_proj`, `down_proj`, `up_proj` |
138
  | **Refinement Stack** | Supervised Style SFT + Binary Anti-Repeat |
139
- | **Final Packaging** | Merged HF safetensors + full GGUF ladder |
140
 
141
  ---
142
 
143
  ## GGUF Quantization Ladder
144
 
145
- All weights are locally hashed. Full recursive validation hashes are in `HASHES.sha256`.
146
-
147
- | Filename | Size | Target Use Case |
148
- |---|---:|---|
149
- | `Z-Image-Engineer-V6-F16.gguf` | 7.498 GiB | Full precision reference. |
150
- | `Z-Image-Engineer-V6-Q8_0.gguf` | 3.986 GiB | Near-lossless; used for local A/B testing. |
151
- | `Z-Image-Engineer-V6-Q6_K.gguf` | 3.079 GiB | High-fidelity balanced footprint. |
152
- | `Z-Image-Engineer-V6-Q5_K_M.gguf` | 2.697 GiB | Daily-driver performance-to-size ratio. |
153
- | `Z-Image-Engineer-V6-Q4_K_M.gguf` | 2.331 GiB | Reliable 4-bit standard. |
154
- | `Z-Image-Engineer-V6-Q3_K_M.gguf` | 1.933 GiB | Lightweight option for tighter setups. |
155
- | `Z-Image-Engineer-V6-MXFP4.gguf` | 2.101 GiB | Alternative compact quantization. |
156
 
157
  ---
158
 
 
13
  - z-image
14
  - z-image-turbo
15
  - qwen3
 
16
  - text-encoder
17
  - comfyui
18
  - lm-studio
 
30
  | **Base Model** | `Tongyi-MAI/Z-Image-Turbo` |
31
  | **Library** | `transformers` |
32
  | **Pipeline Tag** | `text-generation` |
33
+ | **Format** | HF Safetensors |
34
 
35
  ---
36
 
 
38
 
39
  Yes, we jump from V4 to V6. Unlike the usual guy math, this one actually brought the extra two inches.
40
 
41
+ **Z-Image-Engineer V6** is a fine-tuned 4B Qwen text encoder (`Tongyi-MAI/Z-Image-Turbo`) optimized for dual-role performance: a local prompt-enhancement model for LM Studio, and a merged HF text encoder for Z-Image workflows.
42
 
43
  ![Z-Image-Engineer V6 simple A/B with rewrites](evidence/gallery_z_image_engineer_v6_simple_ab_with_rewrites_CONTACT.png)
44
 
 
48
 
49
  V6 transforms minimal seed prompts into rich, highly structured visual narratives. It adds explicit scene composition, lighting direction, material texture, and depth separation while stripping out empty prompt sludge like *"8k, masterpiece, trending on ArtStation."*
50
 
51
+ It can also be used directly as a Z-Image text encoder. This repo contains the merged HF safetensors. The GGUF quantized release lives in the companion repo: [Z-Image-Engineer-V6-GGUF](https://huggingface.co/BennyDaBall/Z-Image-Engineer-V6-GGUF).
52
 
53
  ### Key Use Cases
54
 
55
  - **Prompt Enhancement:** Upgrade simple concepts into descriptive, high-fidelity visual prompts locally.
56
+ - **Text Encoder Swap:** Replace the stock Z-Image Qwen text encoder to generate different conditioning from the same seed.
57
  - **Hybrid Mode:** Use V6 to rewrite your prompt, then use V6 again to encode it. It writes the scene and drives the image model.
58
  - **Private Local Workflow:** Built for LM Studio, ComfyUI, and `llama.cpp`. No API logs, no external telemetry.
59
 
 
88
 
89
  ### LM Studio: Prompt Enhancement
90
 
91
+ Use this merged HF release directly where supported, or download a GGUF quant from [Z-Image-Engineer-V6-GGUF](https://huggingface.co/BennyDaBall/Z-Image-Engineer-V6-GGUF) for LM Studio. No complex system prompt is required.
92
 
93
  ```text
94
  Enhance this image prompt for Z-Image Turbo: a unicorn
 
98
 
99
  ### ComfyUI: Direct Encoder Swap
100
 
101
+ 1. Download a GGUF quant from [Z-Image-Engineer-V6-GGUF](https://huggingface.co/BennyDaBall/Z-Image-Engineer-V6-GGUF).
102
+ 2. Place the GGUF file into `ComfyUI/models/text_encoders/`.
103
+ 3. Add a `CLIPLoaderGGUF` node.
104
+ 4. Set model type to `lumina2`.
105
+ 5. Use it where the stock Z-Image Qwen text encoder would normally go.
106
 
107
  Optional workflow repo:
108
 
 
115
  ```text
116
  UNET: z_image_turbo_bf16.safetensors
117
  VAE: ae.safetensors
118
+ Text Encoder: Z-Image-Engineer-V6-Q8_0.gguf from the GGUF repo
119
  Resolution: 1024x1024
120
  Steps: 8
121
  CFG: 1.0
 
136
  | **Rank / Alpha / Dropout** | 64 / 64 / 0.03 |
137
  | **Target Modules** | `q_proj`, `k_proj`, `v_proj`, `o_proj`, `gate_proj`, `down_proj`, `up_proj` |
138
  | **Refinement Stack** | Supervised Style SFT + Binary Anti-Repeat |
139
+ | **Final Packaging** | Merged HF safetensors |
140
 
141
  ---
142
 
143
  ## GGUF Quantization Ladder
144
 
145
+ The quantized release is separate on purpose:
146
+
147
+ [BennyDaBall/Z-Image-Engineer-V6-GGUF](https://huggingface.co/BennyDaBall/Z-Image-Engineer-V6-GGUF)
148
+
149
+ That repo contains the full GGUF ladder: F16, Q8_0, Q6_K, Q5_K_M, Q4_K_M, Q3_K_M, and MXFP4.
 
 
 
 
 
 
150
 
151
  ---
152
 
RELEASE_MANIFEST.json CHANGED
@@ -2,6 +2,7 @@
2
  "status": "upload_ready_user_approved",
3
  "public_model_name": "Z-Image-Engineer-V6",
4
  "repo_id": "BennyDaBall/Z-Image-Engineer-V6",
 
5
  "base_model": "Tongyi-MAI/Z-Image-Turbo/text_encoder",
6
  "tokenizer": "Tongyi-MAI/Z-Image-Turbo/tokenizer",
7
  "files": [
@@ -63,10 +64,10 @@
63
  },
64
  {
65
  "path": "README.md",
66
- "size_bytes": 7151,
67
  "size_gib": 0,
68
- "sha256": "BF31DA5A1F64F1D7F9AF7C692C82296ED9B2AD59076588BF297E446D6FF54C1C",
69
- "last_write_time": "2026-06-06T01:33:41"
70
  },
71
  {
72
  "path": "tokenizer.json",
@@ -88,107 +89,10 @@
88
  "size_gib": 0,
89
  "sha256": "34126E2486E389F28C11693C2E51641199FB5B53E3E7D6BFA75A6E967C11D3CF",
90
  "last_write_time": "2026-05-30T07:33:27"
91
- },
92
- {
93
- "path": "Z-Image-Engineer-V6-F16.gguf",
94
- "size_bytes": 8051284960,
95
- "size_gib": 7.498,
96
- "sha256": "20DAB6305B76B28808FAD740C7107878DEEC63688E1B318F7BB3A7F707220B0D",
97
- "last_write_time": "2026-06-05T13:36:01"
98
- },
99
- {
100
- "path": "Z-Image-Engineer-V6-MXFP4.gguf",
101
- "size_bytes": 2256005600,
102
- "size_gib": 2.101,
103
- "sha256": "A39695B6714FC4A0A86965F5B2FB8B0CBEF774165EEC8FB9B2379FBEDD86838A",
104
- "last_write_time": "2026-06-05T22:48:03"
105
- },
106
- {
107
- "path": "Z-Image-Engineer-V6-Q3_K_M.gguf",
108
- "size_bytes": 2075617760,
109
- "size_gib": 1.933,
110
- "sha256": "E3F493D971677BA181F67C888AD41E25FD34448BF7EEA03A84F4114EE021B9E3",
111
- "last_write_time": "2026-06-05T22:47:54"
112
- },
113
- {
114
- "path": "Z-Image-Engineer-V6-Q4_K_M.gguf",
115
- "size_bytes": 2503178720,
116
- "size_gib": 2.331,
117
- "sha256": "D666E619EDB2D6DCF2DF013540B22E2592C4FBADB9007B3FB89D4BBE0C4C7C67",
118
- "last_write_time": "2026-06-05T22:47:42"
119
- },
120
- {
121
- "path": "Z-Image-Engineer-V6-Q5_K_M.gguf",
122
- "size_bytes": 2895780320,
123
- "size_gib": 2.697,
124
- "sha256": "0FAB79F032AA34BAAC8607FF8BA720DFB95A0D9A44026DE79288F3FD25A66A05",
125
- "last_write_time": "2026-06-05T22:47:26"
126
- },
127
- {
128
- "path": "Z-Image-Engineer-V6-Q6_K.gguf",
129
- "size_bytes": 3306260960,
130
- "size_gib": 3.079,
131
- "sha256": "A27D6723816462EA1368093A76E9013E996BD4B731EF87327334D50D6DD9534C",
132
- "last_write_time": "2026-06-05T22:47:11"
133
- },
134
- {
135
- "path": "Z-Image-Engineer-V6-Q8_0.gguf",
136
- "size_bytes": 4280404960,
137
- "size_gib": 3.986,
138
- "sha256": "DC4F5476A0F804A7DB73EDA164C0503CDA93858F3EABDE9EA36C68EEDCBA306C",
139
- "last_write_time": "2026-06-05T22:46:50"
140
  }
141
  ],
142
  "ggufs": [
143
- {
144
- "path": "Z-Image-Engineer-V6-F16.gguf",
145
- "size_bytes": 8051284960,
146
- "size_gib": 7.498,
147
- "sha256": "20DAB6305B76B28808FAD740C7107878DEEC63688E1B318F7BB3A7F707220B0D",
148
- "last_write_time": "2026-06-05T13:36:01"
149
- },
150
- {
151
- "path": "Z-Image-Engineer-V6-MXFP4.gguf",
152
- "size_bytes": 2256005600,
153
- "size_gib": 2.101,
154
- "sha256": "A39695B6714FC4A0A86965F5B2FB8B0CBEF774165EEC8FB9B2379FBEDD86838A",
155
- "last_write_time": "2026-06-05T22:48:03"
156
- },
157
- {
158
- "path": "Z-Image-Engineer-V6-Q3_K_M.gguf",
159
- "size_bytes": 2075617760,
160
- "size_gib": 1.933,
161
- "sha256": "E3F493D971677BA181F67C888AD41E25FD34448BF7EEA03A84F4114EE021B9E3",
162
- "last_write_time": "2026-06-05T22:47:54"
163
- },
164
- {
165
- "path": "Z-Image-Engineer-V6-Q4_K_M.gguf",
166
- "size_bytes": 2503178720,
167
- "size_gib": 2.331,
168
- "sha256": "D666E619EDB2D6DCF2DF013540B22E2592C4FBADB9007B3FB89D4BBE0C4C7C67",
169
- "last_write_time": "2026-06-05T22:47:42"
170
- },
171
- {
172
- "path": "Z-Image-Engineer-V6-Q5_K_M.gguf",
173
- "size_bytes": 2895780320,
174
- "size_gib": 2.697,
175
- "sha256": "0FAB79F032AA34BAAC8607FF8BA720DFB95A0D9A44026DE79288F3FD25A66A05",
176
- "last_write_time": "2026-06-05T22:47:26"
177
- },
178
- {
179
- "path": "Z-Image-Engineer-V6-Q6_K.gguf",
180
- "size_bytes": 3306260960,
181
- "size_gib": 3.079,
182
- "sha256": "A27D6723816462EA1368093A76E9013E996BD4B731EF87327334D50D6DD9534C",
183
- "last_write_time": "2026-06-05T22:47:11"
184
- },
185
- {
186
- "path": "Z-Image-Engineer-V6-Q8_0.gguf",
187
- "size_bytes": 4280404960,
188
- "size_gib": 3.986,
189
- "sha256": "DC4F5476A0F804A7DB73EDA164C0503CDA93858F3EABDE9EA36C68EEDCBA306C",
190
- "last_write_time": "2026-06-05T22:46:50"
191
- }
192
  ],
193
  "evidence": [
194
  {
@@ -199,6 +103,6 @@
199
  "last_write_time": "2026-06-05T23:31:26"
200
  }
201
  ],
202
- "generated_at_local": "2026-06-06T01:35:03",
203
  "upload_approved_by_user": true
204
  }
 
2
  "status": "upload_ready_user_approved",
3
  "public_model_name": "Z-Image-Engineer-V6",
4
  "repo_id": "BennyDaBall/Z-Image-Engineer-V6",
5
+ "package_kind": "merged_hf_safetensors",
6
  "base_model": "Tongyi-MAI/Z-Image-Turbo/text_encoder",
7
  "tokenizer": "Tongyi-MAI/Z-Image-Turbo/tokenizer",
8
  "files": [
 
64
  },
65
  {
66
  "path": "README.md",
67
+ "size_bytes": 6895,
68
  "size_gib": 0,
69
+ "sha256": "AF0D74388A53EF4A9B37CF98B05922F1D5D3A888C1C6BBB95D883D89AC423760",
70
+ "last_write_time": "2026-06-06T01:54:57"
71
  },
72
  {
73
  "path": "tokenizer.json",
 
89
  "size_gib": 0,
90
  "sha256": "34126E2486E389F28C11693C2E51641199FB5B53E3E7D6BFA75A6E967C11D3CF",
91
  "last_write_time": "2026-05-30T07:33:27"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
92
  }
93
  ],
94
  "ggufs": [
95
+
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
96
  ],
97
  "evidence": [
98
  {
 
103
  "last_write_time": "2026-06-05T23:31:26"
104
  }
105
  ],
106
+ "generated_at_local": "2026-06-06T01:55:59",
107
  "upload_approved_by_user": true
108
  }