工作流 /
图像 /
ZetaLvX MiniMax H3 Image – ComfyUI Workflow
ZetaLvX MiniMax H3 Image – ComfyUI Workflow
ZetaLvX MiniMax H3 Image – ComfyUI Workflow
A streamlined ComfyUI workflow for running MiniMax H3 Image locally, supporting text-to-image generation, image editing, and multi-reference generation.The workflow includes adjustable aspect ratios, automatic resolution…
图像
19 个节点 203 次使用 Zetalvx 33 KB
需要的权重
这个工作流引用了 1 个权重文件，其中 0 个能在本站模型库里找到。
MiniMax_H3_Image_Zetalvx_ComfyUI_Optimized_Workflow.json
模型库暂未收录
节点构成
MarkdownNote
× 4
PrimitiveFloat
× 2
ComfyMathExpression
× 2
ResolutionSelector
× 1
UNETLoader
× 1
CLIPLoader
× 1
VAELoader
× 1
MiniMaxH3ImageConditioning
× 1
ConditioningZeroOut
× 1
KSampler
× 1
VAEDecode
× 1
PreviewImage
× 1
SaveImage
× 1
LoadImage
× 1
官方标注的模型
minimax_h3_image_pruned_int8_convrot.safetensorsminimax_h3_image_vae_fp16.safetensorsqwen3vl_32b_minimax_h3_int8_convrot.safetensors
工作流 JSON
和上面「下载工作流 JSON」是同一份文件。
1{
2 "id": "d8dd10fc-d3e6-4d36-8c45-01e2b625b02c",
3 "revision": 1,
4 "last_node_id": 20,
5 "last_link_id": 22,
6 "nodes": [
7 {
8 "id": 1,
9 "type": "ResolutionSelector",
10 "pos": [
11 -1250,
12 -600
13 ],
14 "size": [
15 410,
16 174
17 ],
18 "flags": [],
19 "order": 10,
20 "mode": 0,
21 "showAdvanced": false,
22 "inputs": [
23 {
24 "name": "megapixels",
25 "type": "FLOAT",
26 "widget": {
27 "name": "megapixels"
28 },
29 "link": 22
30 }
31 ],
32 "outputs": [
33 {
34 "name": "width",
35 "type": "INT",
36 "links": [
37 1,
38 4
39 ]
40 },
41 {
42 "name": "height",
43 "type": "INT",
44 "links": [
45 2,
46 5
47 ]
48 }
49 ],
50 "title": "01 | ASPECT RATIO (change this)",
51 "properties": {
52 "cnr_id": "comfy-core",
53 "ver": "0.45.0",
54 "Node name for S&R": "ResolutionSelector"
55 },
56 "widgets_values": [
57 "16:9 (Widescreen)",
58 16,
59 32
60 ],
61 "widgets_values_named": {
62 "aspect_ratio": "16:9 (Widescreen)",
63 "megapixels": 16,
64 "multiple": 32
65 }
66 },
67 {
68 "id": 2,
69 "type": "PrimitiveFloat",
70 "pos": [
71 -1250,
72 -330
73 ],
74 "size": [
75 410,
76 92
77 ],
78 "flags": [],
79 "order": 0,
80 "mode": 0,
81 "inputs": [],
82 "outputs": [
83 {
84 "name": "FLOAT",
85 "type": "FLOAT",
86 "links": [
87 3,
88 6
89 ]
90 }
91 ],
92 "title": "02 | TARGET MEGAPIXELS (4–20)",
93 "properties": {
94 "cnr_id": "comfy-core",
95 "ver": "0.45.0",
96 "Node name for S&R": "PrimitiveFloat"
97 },
98 "widgets_values": [
99 20
100 ],
101 "widgets_values_named": {
102 "value": 20
103 }
104 },
105 {
106 "id": 3,
107 "type": "ComfyMathExpression",
108 "pos": [
109 -770,
110 -600
111 ],
112 "size": [
113 440,
114 182
115 ],
116 "flags": [],
117 "order": 11,
118 "mode": 0,
119 "inputs": [
120 {
121 "name": "values.a",
122 "type": "FLOAT,INT,BOOLEAN",
123 "link": 1
124 },
125 {
126 "name": "values.b",
127 "shape": 7,
128 "type": "FLOAT,INT,BOOLEAN",
129 "link": 2
130 },
131 {
132 "name": "values.c",
133 "shape": 7,
134 "type": "FLOAT,INT,BOOLEAN",
135 "link": 3
136 },
137 {
138 "name": "values.d",
139 "shape": 7,
140 "type": "FLOAT,INT,BOOLEAN",
141 "link": null
142 }
143 ],
144 "outputs": [
145 {
146 "name": "FLOAT",
147 "type": "FLOAT",
148 "links": null
149 },
150 {
151 "name": "INT",
152 "type": "INT",
153 "links": [
154 7
155 ]
156 },
157 {
158 "name": "BOOL",
159 "type": "BOOLEAN",
160 "links": null
161 }
162 ],
163 "title": "Calculated width",
164 "properties": {
165 "cnr_id": "comfy-core",
166 "ver": "0.45.0",
167 "Node name for S&R": "ComfyMathExpression"
168 },
169 "widgets_values": [
170 "round(sqrt(a / b * c * 1000000) / 32) * 32"
171 ],
172 "widgets_values_named": {
173 "expression": "round(sqrt(a / b * c * 1000000) / 32) * 32"
174 }
175 },
176 {
177 "id": 4,
178 "type": "ComfyMathExpression",
179 "pos": [
180 -770,
181 -330
182 ],
183 "size": [
184 440,
185 182
186 ],
187 "flags": [],
188 "order": 12,
189 "mode": 0,
190 "inputs": [
191 {
192 "name": "values.a",
193 "type": "FLOAT,INT,BOOLEAN",
194 "link": 4
195 },
196 {
197 "name": "values.b",
198 "shape": 7,
199 "type": "FLOAT,INT,BOOLEAN",
200 "link": 5
201 },
202 {
203 "name": "values.c",
204 "shape": 7,
205 "type": "FLOAT,INT,BOOLEAN",
206 "link": 6
207 },
208 {
209 "name": "values.d",
210 "shape": 7,
211 "type": "FLOAT,INT,BOOLEAN",
212 "link": null
213 }
214 ],
215 "outputs": [
216 {
217 "name": "FLOAT",
218 "type": "FLOAT",
219 "links": null
220 },
221 {
222 "name": "INT",
223 "type": "INT",
224 "links": [
225 8
226 ]
227 },
228 {
229 "name": "BOOL",
230 "type": "BOOLEAN",
231 "links": null
232 }
233 ],
234 "title": "Calculated height",
235 "properties": {
236 "cnr_id": "comfy-core",
237 "ver": "0.45.0",
238 "Node name for S&R": "ComfyMathExpression"
239 },
240 "widgets_values": [
241 "round(sqrt(b / a * c * 1000000) / 32) * 32"
242 ],
243 "widgets_values_named": {
244 "expression": "round(sqrt(b / a * c * 1000000) / 32) * 32"
245 }
246 },
247 {
248 "id": 5,
249 "type": "UNETLoader",
250 "pos": [
251 -225,
252 -595
253 ],
254 "size": [
255 530,
256 94
257 ],
258 "flags": [],
259 "order": 1,
260 "mode": 0,
261 "inputs": [],
262 "outputs": [
263 {
264 "name": "MODEL",
265 "type": "MODEL",
266 "links": [
267 11
268 ]
269 }
270 ],
271 "title": "ZetaLvX • MiniMax H3 • diffusion model",
272 "properties": {
273 "cnr_id": "comfy-core",
274 "ver": "0.45.0",
275 "Node name for S&R": "UNETLoader"
276 },
277 "widgets_values": [
278 "minimax_h3_image_pruned_int8_convrot.safetensors",
279 "default"
280 ],
281 "widgets_values_named": {
282 "unet_name": "minimax_h3_image_pruned_int8_convrot.safetensors",
283 "weight_dtype": "default"
284 },
285 "color": "#223",
286 "bgcolor": "#335"
287 },
288 {
289 "id": 6,
290 "type": "CLIPLoader",
291 "pos": [
292 -225,
293 -435
294 ],
295 "size": [
296 530,
297 128
298 ],
299 "flags": [],
300 "order": 2,
301 "mode": 0,
302 "inputs": [],
303 "outputs": [
304 {
305 "name": "CLIP",
306 "type": "CLIP",
307 "links": [
308 9
309 ]
310 }
311 ],
312 "title": "ZetaLvX • MiniMax H3 • text encoder",
313 "properties": {
314 "cnr_id": "comfy-core",
315 "ver": "0.45.0",
316 "Node name for S&R": "CLIPLoader"
317 },
318 "widgets_values": [
319 "qwen3vl_32b_minimax_h3_int8_convrot.safetensors",
320 "minimax",
321 "default"
322 ],
323 "widgets_values_named": {
324 "clip_name": "qwen3vl_32b_minimax_h3_int8_convrot.safetensors",
325 "type": "minimax",
326 "device": "default"
327 }
328 },
329 {
330 "id": 7,
331 "type": "VAELoader",
332 "pos": [
333 -225,
334 -245
335 ],
336 "size": [
337 530,
338 90
339 ],
340 "flags": [],
341 "order": 3,
342 "mode": 0,
343 "inputs": [],
344 "outputs": [
345 {
346 "name": "VAE",
347 "type": "VAE",
348 "links": [
349 10,
350 17
351 ]
352 }
353 ],
354 "title": "ZetaLvX • MiniMax H3 • VAE",
355 "properties": {
356 "cnr_id": "comfy-core",
357 "ver": "0.45.0",
358 "Node name for S&R": "VAELoader"
359 },
360 "widgets_values": [
361 "minimax_h3_image_vae_fp16.safetensors"
362 ],
363 "widgets_values_named": {
364 "vae_name": "minimax_h3_image_vae_fp16.safetensors"
365 },
366 "color": "#322",
367 "bgcolor": "#533"
368 },
369 {
370 "id": 8,
371 "type": "MiniMaxH3ImageConditioning",
372 "pos": [
373 381.22582806043386,
374 -608.8751245924263
375 ],
376 "size": [
377 666.8817693536927,
378 765
379 ],
380 "flags": [],
381 "order": 13,
382 "mode": 0,
383 "inputs": [
384 {
385 "name": "clip",
386 "type": "CLIP",
387 "link": 9
388 },
389 {
390 "name": "vae",
391 "shape": 7,
392 "type": "VAE",
393 "link": 10
394 },
395 {
396 "label": "image_0",
397 "name": "images.image_0",
398 "shape": 7,
399 "type": "IMAGE",
400 "link": 21
401 },
402 {
403 "name": "images.image_1",
404 "shape": 7,
405 "type": "IMAGE",
406 "link": null
407 },
408 {
409 "name": "width",
410 "type": "INT",
411 "widget": {
412 "name": "width"
413 },
414 "link": 7
415 },
416 {
417 "name": "height",
418 "type": "INT",
419 "widget": {
420 "name": "height"
421 },
422 "link": 8
423 },
424 {
425 "label": "image_2",
426 "name": "images.image_2",
427 "shape": 7,
428 "type": "IMAGE",
429 "link": null
430 },
431 {
432 "label": "image_3",
433 "name": "images.image_3",
434 "shape": 7,
435 "type": "IMAGE",
436 "link": null
437 },
438 {
439 "label": "image_4",
440 "name": "images.image_4",
441 "shape": 7,
442 "type": "IMAGE",
443 "link": null
444 },
445 {
446 "label": "image_5",
447 "name": "images.image_5",
448 "shape": 7,
449 "type": "IMAGE",
450 "link": null
451 },
452 {
453 "label": "image_6",
454 "name": "images.image_6",
455 "shape": 7,
456 "type": "IMAGE",
457 "link": null
458 },
459 {
460 "label": "image_7",
461 "name": "images.image_7",
462 "shape": 7,
463 "type": "IMAGE",
464 "link": null
465 },
466 {
467 "label": "image_8",
468 "name": "images.image_8",
469 "shape": 7,
470 "type": "IMAGE",
471 "link": null
472 },
473 {
474 "label": "image_9",
475 "name": "images.image_9",
476 "shape": 7,
477 "type": "IMAGE",
478 "link": null
479 },
480 {
481 "label": "image_10",
482 "name": "images.image_10",
483 "shape": 7,
484 "type": "IMAGE",
485 "link": null
486 },
487 {
488 "label": "image_11",
489 "name": "images.image_11",
490 "shape": 7,
491 "type": "IMAGE",
492 "link": null
493 },
494 {
495 "label": "image_12",
496 "name": "images.image_12",
497 "shape": 7,
498 "type": "IMAGE",
499 "link": null
500 },
501 {
502 "label": "image_13",
503 "name": "images.image_13",
504 "shape": 7,
505 "type": "IMAGE",
506 "link": null
507 },
508 {
509 "label": "image_14",
510 "name": "images.image_14",
511 "shape": 7,
512 "type": "IMAGE",
513 "link": null
514 },
515 {
516 "label": "image_15",
517 "name": "images.image_15",
518 "shape": 7,
519 "type": "IMAGE",
520 "link": null
521 }
522 ],
523 "outputs": [
524 {
525 "name": "positive",
526 "type": "CONDITIONING",
527 "links": [
528 12,
529 13
530 ]
531 },
532 {
533 "name": "LATENT",
534 "type": "LATENT",
535 "links": [
536 15
537 ]
538 }
539 ],
540 "title": "PROMPT • MiniMax H3 Image",
541 "properties": {
542 "cnr_id": "comfy-core",
543 "ver": "0.45.0",
544 "Node name for S&R": "MiniMaxH3ImageConditioning"
545 },
546 "widgets_values": [
547 "Edit the provided image_1 by ADDING TYPOGRAPHY ONLY. This is a strict image preservation task.\n\nKeep the original photograph completely unchanged. Preserve the exact same person, facial features, expression, hairstyle, sunglasses, jacket, colors, lighting, geometric background, framing, camera angle, and image proportions. Do not crop, extend, redraw, retouch, enhance, recolor, or regenerate any part of the original image.\n\nCreate an elegant, high-impact YouTube thumbnail text overlay using these EXACT words:\n\nH3-IMAGE\nCOMFYUI\nOPTIMIZED WORKFLOWS\n\nPlace the typography mainly across the lower third of the image, avoiding the woman's eyes, glasses, nose, and mouth.\n\n\"H3-IMAGE\" is the dominant headline: very large, bold, modern white sans-serif typography, with \"H3\" highlighted in bright cyan.\n\n\"COMFYUI\" is a smaller secondary heading in white.\n\n\"OPTIMIZED WORKFLOWS\" is a compact uppercase subtitle in cyan, positioned below the main heading.\n\nUse precise typography, consistent letter spacing, sharp edges, professional graphic design, and excellent readability at small thumbnail size.\n\nAdd only a subtle dark translucent gradient or soft shadow directly behind the text if necessary for contrast. Do not otherwise modify the image.\n\nKeep every letter fully visible inside the image boundaries with safe margins. Avoid overlapping the woman's face. Do not add any logos, icons, symbols, watermarks, extra words, frames, decorative graphics, or AI-themed elements.\n\nThe final image must be the original photograph with only the requested professionally designed text overlay added.",
548 4608,
549 3456
550 ],
551 "widgets_values_named": {
552 "prompt": "Edit the provided image_1 by ADDING TYPOGRAPHY ONLY. This is a strict image preservation task.\n\nKeep the original photograph completely unchanged. Preserve the exact same person, facial features, expression, hairstyle, sunglasses, jacket, colors, lighting, geometric background, framing, camera angle, and image proportions. Do not crop, extend, redraw, retouch, enhance, recolor, or regenerate any part of the original image.\n\nCreate an elegant, high-impact YouTube thumbnail text overlay using these EXACT words:\n\nH3-IMAGE\nCOMFYUI\nOPTIMIZED WORKFLOWS\n\nPlace the typography mainly across the lower third of the image, avoiding the woman's eyes, glasses, nose, and mouth.\n\n\"H3-IMAGE\" is the dominant headline: very large, bold, modern white sans-serif typography, with \"H3\" highlighted in bright cyan.\n\n\"COMFYUI\" is a smaller secondary heading in white.\n\n\"OPTIMIZED WORKFLOWS\" is a compact uppercase subtitle in cyan, positioned below the main heading.\n\nUse precise typography, consistent letter spacing, sharp edges, professional graphic design, and excellent readability at small thumbnail size.\n\nAdd only a subtle dark translucent gradient or soft shadow directly behind the text if necessary for contrast. Do not otherwise modify the image.\n\nKeep every letter fully visible inside the image boundaries with safe margins. Avoid overlapping the woman's face. Do not add any logos, icons, symbols, watermarks, extra words, frames, decorative graphics, or AI-themed elements.\n\nThe final image must be the original photograph with only the requested professionally designed text overlay added.",
553 "width": 4608,
554 "height": 3456
555 }
556 },
557 {
558 "id": 9,
559 "type": "ConditioningZeroOut",
560 "pos": [
561 1117.6388972349741,
562 -531.9444806833794
563 ],
564 "size": [
565 320,
566 78
567 ],
568 "flags": [],
569 "order": 14,
570 "mode": 0,
571 "inputs": [
572 {
573 "name": "conditioning",
574 "type": "CONDITIONING",
575 "link": 13
576 }
577 ],
578 "outputs": [
579 {
580 "name": "CONDITIONING",
581 "type": "CONDITIONING",
582 "links": [
583 14
584 ]
585 }
586 ],
587 "title": "No CFG • empty negative",
588 "properties": {
589 "cnr_id": "comfy-core",
590 "ver": "0.45.0",
591 "Node name for S&R": "ConditioningZeroOut"
592 }
593 },
594 {
595 "id": 10,
596 "type": "KSampler",
597 "pos": [
598 1129.428635689974,
599 -365.23985297537945
600 ],
601 "size": [
602 510,
603 350
604 ],
605 "flags": [],
606 "order": 15,
607 "mode": 0,
608 "inputs": [
609 {
610 "name": "model",
611 "type": "MODEL",
612 "link": 11
613 },
614 {
615 "name": "positive",
616 "type": "CONDITIONING",
617 "link": 12
618 },
619 {
620 "name": "negative",
621 "type": "CONDITIONING",
622 "link": 14
623 },
624 {
625 "name": "latent_image",
626 "type": "LATENT",
627 "link": 15
628 }
629 ],
630 "outputs": [
631 {
632 "name": "LATENT",
633 "type": "LATENT",
634 "links": [
635 16
636 ]
637 }
638 ],
639 "title": "SAMPLING • 6 steps / Euler",
640 "properties": {
641 "cnr_id": "comfy-core",
642 "ver": "0.45.0",
643 "Node name for S&R": "KSampler"
644 },
645 "widgets_values": [
646 77,
647 "fixed",
648 6,
649 1,
650 "euler",
651 "simple",
652 1
653 ],
654 "widgets_values_named": {
655 "seed": 77,
656 "control_after_generate": "fixed",
657 "steps": 6,
658 "cfg": 1,
659 "sampler_name": "euler",
660 "scheduler": "simple",
661 "denoise": 1
662 }
663 },
664 {
665 "id": 11,
666 "type": "VAEDecode",
667 "pos": [
668 1736.1108011690271,
669 -587.2915900157541
670 ],
671 "size": [
672 310,
673 80
674 ],
675 "flags": [],
676 "order": 16,
677 "mode": 0,
678 "inputs": [
679 {
680 "name": "samples",
681 "type": "LATENT",
682 "link": 16
683 },
684 {
685 "name": "vae",
686 "type": "VAE",
687 "link": 17
688 }
689 ],
690 "outputs": [
691 {
692 "name": "IMAGE",
693 "type": "IMAGE",
694 "links": [
695 18,
696 19
697 ]
698 }
699 ],
700 "title": "DECODE • VAE FP16",
701 "properties": {
702 "cnr_id": "comfy-core",
703 "ver": "0.45.0",
704 "Node name for S&R": "VAEDecode"
705 }
706 },
707 {
708 "id": 12,
709 "type": "PreviewImage",
710 "pos": [
711 2146.1108011690267,
712 -637.2915900157541
713 ],
714 "size": [
715 600,
716 460
717 ],
718 "flags": [],
719 "order": 17,
720 "mode": 0,
721 "inputs": [
722 {
723 "name": "images",
724 "type": "IMAGE",
725 "link": 18
726 }
727 ],
728 "outputs": [
729 {
730 "name": "images",
731 "type": "IMAGE",
732 "links": null
733 }
734 ],
735 "title": "Image preview",
736 "properties": {
737 "cnr_id": "comfy-core",
738 "ver": "0.45.0",
739 "Node name for S&R": "PreviewImage"
740 }
741 },
742 {
743 "id": 13,
744 "type": "SaveImage",
745 "pos": [
746 1756.3673811690269,
747 -99.4568378457679
748 ],
749 "size": [
750 500,
751 120
752 ],
753 "flags": [],
754 "order": 18,
755 "mode": 0,
756 "inputs": [
757 {
758 "name": "images",
759 "type": "IMAGE",
760 "link": 19
761 }
762 ],
763 "outputs": [
764 {
765 "name": "images",
766 "type": "IMAGE",
767 "links": null
768 }
769 ],
770 "title": "Save PNG",
771 "properties": {
772 "cnr_id": "comfy-core",
773 "ver": "0.45.0",
774 "Node name for S&R": "SaveImage"
775 },
776 "widgets_values": [
777 "ZetaLvX_MiniMax_H3_Image"
778 ],
779 "widgets_values_named": {
780 "filename_prefix": "ZetaLvX_MiniMax_H3_Image"
781 }
782 },
783 {
784 "id": 14,
785 "type": "MarkdownNote",
786 "pos": [
787 -976.0164923235902,
788 626.1580569320458
789 ],
790 "size": [
791 740,
792 750
793 ],
794 "flags": [],
795 "order": 4,
796 "mode": 0,
797 "inputs": [],
798 "outputs": [],
799 "title": "GUIDE / USAGE",
800 "properties": {
801 "Node name for S&R": "MarkdownNote",
802 "cnr_id": "comfy-core",
803 "ver": "0.45.0"
804 },
805 "widgets_values": [
806 "# ZetaLvX | MiniMax H3 Image\n### Text-to-image · Image editing · Multi-reference\n\nThe same model supports image generation and image editing. Connect reference images only when needed.\n\n**Text-to-image**\n1. Keep all `image_0`–`image_15` inputs disconnected.\n2. Enter a generation prompt in **PROMPT**.\n3. Set the aspect ratio and target megapixels, then run the workflow.\n\n**Image editing**\n1. Add a built-in **Load Image** node and connect its IMAGE output to `image_0`.\n2. Describe the requested edit and what should remain unchanged.\n3. Refer to the source image as `<Picture 1>` in the prompt.\n\n**Multi-reference generation or editing**\n1. Connect additional Load Image nodes to `image_1`, `image_2`, etc.\n2. Refer to the connected pictures in the prompt as `<Picture 1>`, `<Picture 2>`, etc.\n3. Describe each image's purpose, such as subject, outfit, composition, or background.\n\n**Settings**\n- Aspect-ratio selector with adjustable 4–20 MP target resolution.\n- Sampler: **Euler**; scheduler: **Simple**; **6 steps**, **CFG 1**, **denoise 1**.\n- Prompt enhancement is not included.\n- Connect the VAE for editing so the reference images can also condition the latent representation.\n\n**Links**\n- YouTube: https://youtube.com/@zetalvx\n- Patreon: https://www.patreon.com/Zetalvx\n- GitHub: https://github.com/ZETALVX"
807 ],
808 "widgets_values_named": {
809 "text": "# ZetaLvX | MiniMax H3 Image\n### Text-to-image · Image editing · Multi-reference\n\nThe same model supports image generation and image editing. Connect reference images only when needed.\n\n**Text-to-image**\n1. Keep all `image_0`–`image_15` inputs disconnected.\n2. Enter a generation prompt in **PROMPT**.\n3. Set the aspect ratio and target megapixels, then run the workflow.\n\n**Image editing**\n1. Add a built-in **Load Image** node and connect its IMAGE output to `image_0`.\n2. Describe the requested edit and what should remain unchanged.\n3. Refer to the source image as `<Picture 1>` in the prompt.\n\n**Multi-reference generation or editing**\n1. Connect additional Load Image nodes to `image_1`, `image_2`, etc.\n2. Refer to the connected pictures in the prompt as `<Picture 1>`, `<Picture 2>`, etc.\n3. Describe each image's purpose, such as subject, outfit, composition, or background.\n\n**Settings**\n- Aspect-ratio selector with adjustable 4–20 MP target resolution.\n- Sampler: **Euler**; scheduler: **Simple**; **6 steps**, **CFG 1**, **denoise 1**.\n- Prompt enhancement is not included.\n- Connect the VAE for editing so the reference images can also condition the latent representation.\n\n**Links**\n- YouTube: https://youtube.com/@zetalvx\n- Patreon: https://www.patreon.com/Zetalvx\n- GitHub: https://github.com/ZETALVX"
810 },
811 "color": "#432",
812 "bgcolor": "#653"
813 },
814 {
815 "id": 15,
816 "type": "MarkdownNote",
817 "pos": [
818 -169.6878897987749,
819 405.6601065548359
820 ],
821 "size": [
822 540,
823 540
824 ],
825 "flags": [],
826 "order": 5,
827 "mode": 0,
828 "inputs": [],
829 "outputs": [],
830 "title": "MODEL FILES",
831 "properties": {
832 "Node name for S&R": "MarkdownNote",
833 "cnr_id": "comfy-core",
834 "ver": "0.45.0"
835 },
836 "widgets_values": [
837 "# Model files\n**Diffusion model**\n`minimax_h3_image_pruned_int8_convrot.safetensors`\n\n**Text encoder**\n`qwen3vl_32b_minimax_h3_int8_convrot.safetensors`\n\n**VAE**\n`minimax_h3_image_vae_fp16.safetensors`\n\nThe same model configuration is used for text-to-image and image editing. Image references are optional.\n\n### Prompt enhancement\nAn additional prompt-enhancement model is not required by this workflow. You can write a detailed prompt directly in the conditioning node."
838 ],
839 "widgets_values_named": {
840 "text": "# Model files\n**Diffusion model**\n`minimax_h3_image_pruned_int8_convrot.safetensors`\n\n**Text encoder**\n`qwen3vl_32b_minimax_h3_int8_convrot.safetensors`\n\n**VAE**\n`minimax_h3_image_vae_fp16.safetensors`\n\nThe same model configuration is used for text-to-image and image editing. Image references are optional.\n\n### Prompt enhancement\nAn additional prompt-enhancement model is not required by this workflow. You can write a detailed prompt directly in the conditioning node."
841 },
842 "color": "#432",
843 "bgcolor": "#653"
844 },
845 {
846 "id": 16,
847 "type": "MarkdownNote",
848 "pos": [
849 1091.7141704423227,
850 205.2204328980684
851 ],
852 "size": [
853 593.2988782180726,
854 526.944215146494
855 ],
856 "flags": [],
857 "order": 6,
858 "mode": 0,
859 "inputs": [],
860 "outputs": [],
861 "title": "RESOLUTION / QUALITY",
862 "properties": {
863 "Node name for S&R": "MarkdownNote",
864 "cnr_id": "comfy-core",
865 "ver": "0.45.0"
866 },
867 "widgets_values": [
868 "# Resolution and quality\nThe model renders directly at the requested resolution: **no upscaler is used**.\n\nReference ~16–17 MP sizes published by MiniMax:\n- Square: `4096 × 4096`\n- 4:3: `4736 × 3552`\n- 3:2: `5024 × 3360`\n- 16:9: `5472 × 3072`\n- 21:9: `6272 × 2688`\n\nThe dynamic control uses **exact decimal megapixels** (1 MP = 1,000,000 pixels), so it does not try to force the exact preset dimensions from the official table.\n\nFor clean comparisons: keep the same seed, prompt, steps, and sampler.\n\nSteps are **6** because the model is distilled and does not use a classic negative prompt or standard CFG workflow.\n"
869 ],
870 "widgets_values_named": {
871 "text": "# Resolution and quality\nThe model renders directly at the requested resolution: **no upscaler is used**.\n\nReference ~16–17 MP sizes published by MiniMax:\n- Square: `4096 × 4096`\n- 4:3: `4736 × 3552`\n- 3:2: `5024 × 3360`\n- 16:9: `5472 × 3072`\n- 21:9: `6272 × 2688`\n\nThe dynamic control uses **exact decimal megapixels** (1 MP = 1,000,000 pixels), so it does not try to force the exact preset dimensions from the official table.\n\nFor clean comparisons: keep the same seed, prompt, steps, and sampler.\n\nSteps are **6** because the model is distilled and does not use a classic negative prompt or standard CFG workflow.\n"
872 },
873 "color": "#432",
874 "bgcolor": "#653"
875 },
876 {
877 "id": 18,
878 "type": "MarkdownNote",
879 "pos": [
880 436.3744084295813,
881 278.52936169189866
882 ],
883 "size": [
884 600,
885 570
886 ],
887 "flags": [],
888 "order": 7,
889 "mode": 0,
890 "inputs": [],
891 "outputs": [],
892 "title": "IMAGE INPUTS / PROMPT REFERENCES",
893 "properties": {
894 "Node name for S&R": "MarkdownNote"
895 },
896 "widgets_values": [
897 "# Image inputs and prompt references\n\nThe inputs on the ComfyUI node are named `image_0` through `image_15`. MiniMax H3 Image identifies connected pictures in the text prompt using **`<Picture 1>`, `<Picture 2>`, ...**.\n\n| ComfyUI input | Prompt reference when connected in order |\n|---|---|\n| `image_0` | `<Picture 1>` |\n| `image_1` | `<Picture 2>` |\n| `image_2` | `<Picture 3>` |\n| ... | ... |\n| `image_15` | `<Picture 16>` |\n\n**The numbering in the prompt follows the order of connected images.** Empty inputs are skipped. For predictable references, connect images consecutively from `image_0`.\n\n### Prompt examples\n**Editing one image:**\n`Edit <Picture 1> by changing the jacket to red. Preserve the subject's identity, face, pose, and background.`\n\n**Combining references:**\n`Use <Picture 1> for the subject, <Picture 2> for the outfit, and <Picture 3> for the background. Create a realistic cinematic photograph.`\n\n**Without image inputs:**\nWrite a normal text-to-image prompt; do not include picture references.\n\nThe 16 input slots are optional. Additional connected references may increase memory usage and processing time.\n"
898 ],
899 "widgets_values_named": {
900 "text": "# Image inputs and prompt references\n\nThe inputs on the ComfyUI node are named `image_0` through `image_15`. MiniMax H3 Image identifies connected pictures in the text prompt using **`<Picture 1>`, `<Picture 2>`, ...**.\n\n| ComfyUI input | Prompt reference when connected in order |\n|---|---|\n| `image_0` | `<Picture 1>` |\n| `image_1` | `<Picture 2>` |\n| `image_2` | `<Picture 3>` |\n| ... | ... |\n| `image_15` | `<Picture 16>` |\n\n**The numbering in the prompt follows the order of connected images.** Empty inputs are skipped. For predictable references, connect images consecutively from `image_0`.\n\n### Prompt examples\n**Editing one image:**\n`Edit <Picture 1> by changing the jacket to red. Preserve the subject's identity, face, pose, and background.`\n\n**Combining references:**\n`Use <Picture 1> for the subject, <Picture 2> for the outfit, and <Picture 3> for the background. Create a realistic cinematic photograph.`\n\n**Without image inputs:**\nWrite a normal text-to-image prompt; do not include picture references.\n\nThe 16 input slots are optional. Additional connected references may increase memory usage and processing time.\n"
901 },
902 "color": "#432",
903 "bgcolor": "#653"
904 },
905 {
906 "id": 19,
907 "type": "LoadImage",
908 "pos": [
909 -1264.0962051695783,
910 -28.351779567149833
911 ],
912 "size": [
913 966.6932844716769,
914 569.2047913305175
915 ],
916 "flags": [],
917 "order": 8,
918 "mode": 0,
919 "inputs": [],
920 "outputs": [
921 {
922 "name": "IMAGE",
923 "type": "IMAGE",
924 "links": [
925 21
926 ]
927 },
928 {
929 "name": "MASK",
930 "type": "MASK",
931 "links": []
932 }
933 ],
934 "properties": {
935 "cnr_id": "comfy-core",
936 "ver": "0.45.0",
937 "Node name for S&R": "LoadImage"
938 },
939 "widgets_values": [
940 "ZetaLvX_MiniMax_H3_T2I_00029_.png [output]",
941 "image"
942 ],
943 "widgets_values_named": {
944 "image": "ZetaLvX_MiniMax_H3_T2I_00029_.png [output]",
945 "upload": "image"
946 }
947 },
948 {
949 "id": 20,
950 "type": "PrimitiveFloat",
951 "pos": [
952 -1265.7044263916016,
953 -690.0144467010498
954 ],
955 "size": [
956 250,
957 92
958 ],
959 "flags": {
960 "collapsed": true
961 },
962 "order": 9,
963 "mode": 0,
964 "inputs": [],
965 "outputs": [
966 {
967 "name": "FLOAT",
968 "type": "FLOAT",
969 "links": [
970 22
971 ]
972 }
973 ],
974 "title": "Fixed ratio reference (internal)",
975 "properties": {
976 "cnr_id": "comfy-core",
977 "ver": "0.45.0",
978 "Node name for S&R": "PrimitiveFloat"
979 },
980 "widgets_values": [
981 16
982 ],
983 "widgets_values_named": {
984 "value": 16
985 }
986 }
987 ],
988 "links": [
989 [
990 11,
991 5,
992 0,
993 10,
994 0,
995 "MODEL"
996 ],
997 [
998 12,
999 8,
1000 0,
1001 10,
1002 1,
1003 "CONDITIONING"
1004 ],
1005 [
1006 13,
1007 8,
1008 0,
1009 9,
1010 0,
1011 "CONDITIONING"
1012 ],
1013 [
1014 14,
1015 9,
1016 0,
1017 10,
1018 2,
1019 "CONDITIONING"
1020 ],
1021 [
1022 15,
1023 8,
1024 1,
1025 10,
1026 3,
1027 "LATENT"
1028 ],
1029 [
1030 16,
1031 10,
1032 0,
1033 11,
1034 0,
1035 "LATENT"
1036 ],
1037 [
1038 17,
1039 7,
1040 0,
1041 11,
1042 1,
1043 "VAE"
1044 ],
1045 [
1046 18,
1047 11,
1048 0,
1049 12,
1050 0,
1051 "IMAGE"
1052 ],
1053 [
1054 19,
1055 11,
1056 0,
1057 13,
1058 0,
1059 "IMAGE"
1060 ],
1061 [
1062 9,
1063 6,
1064 0,
1065 8,
1066 0,
1067 "CLIP"
1068 ],
1069 [
1070 10,
1071 7,
1072 0,
1073 8,
1074 1,
1075 "VAE"
1076 ],
1077 [
1078 21,
1079 19,
1080 0,
1081 8,
1082 2,
1083 "IMAGE"
1084 ],
1085 [
1086 7,
1087 3,
1088 1,
1089 8,
1090 4,
1091 "INT"
1092 ],
1093 [
1094 8,
1095 4,
1096 1,
1097 8,
1098 5,
1099 "INT"
1100 ],
1101 [
1102 22,
1103 20,
1104 0,
1105 1,
1106 0,
1107 "FLOAT"
1108 ],
1109 [
1110 1,
1111 1,
1112 0,
1113 3,
1114 0,
1115 "INT"
1116 ],
1117 [
1118 2,
1119 1,
1120 1,
1121 3,
1122 1,
1123 "INT"
1124 ],
1125 [
1126 3,
1127 2,
1128 0,
1129 3,
1130 2,
1131 "FLOAT"
1132 ],
1133 [
1134 4,
1135 1,
1136 0,
1137 4,
1138 0,
1139 "INT"
1140 ],
1141 [
1142 5,
1143 1,
1144 1,
1145 4,
1146 1,
1147 "INT"
1148 ],
1149 [
1150 6,
1151 2,
1152 0,
1153 4,
1154 2,
1155 "FLOAT"
1156 ]
1157 ],
1158 "groups": [
1159 {
1160 "id": 1,
1161 "title": "1 — FORMATO e MP",
1162 "bounding": [
1163 -1270,
1164 -675,
1165 970.1579103452759,
1166 582.0966430544435
1167 ],
1168 "color": "#596f95",
1169 "flags": []
1170 },
1171 {
1172 "id": 2,
1173 "title": "2 — MODELLI",
1174 "bounding": [
1175 -250,
1176 -675,
1177 585,
1178 748
1179 ],
1180 "color": "#c28a54",
1181 "flags": []
1182 },
1183 {
1184 "id": 3,
1185 "title": "3 — PROMPT e SIZE",
1186 "bounding": [
1187 370,
1188 -675,
1189 680,
1190 768
1191 ],
1192 "color": "#5b9d6c",
1193 "flags": []
1194 },
1195 {
1196 "id": 4,
1197 "title": "4 — SAMPLING",
1198 "bounding": [
1199 1092.6388972349741,
1200 -651.9444806833794,
1201 565,
1202 738
1203 ],
1204 "color": "#8c6286",
1205 "flags": []
1206 },
1207 {
1208 "id": 5,
1209 "title": "5 — OUTPUT",
1210 "bounding": [
1211 1706.1108011690271,
1212 -692.2915900157541,
1213 1089.7606070922848,
1214 912.846
1215 ],
1216 "color": "#52a2a5",
1217 "flags": []
1218 }
1219 ],
1220 "config": [],
1221 "extra": {
1222 "ds": {
1223 "scale": 0.513158118230707,
1224 "offset": [
1225 1680.5809393388286,
1226 1060.2177060504428
1227 ]
1228 },
1229 "frontendVersion": "1.53.6",
1230 "VHS_latentpreview": false,
1231 "VHS_latentpreviewrate": 0,
1232 "VHS_MetadataImage": true,
1233 "VHS_KeepIntermediate": true
1234 },
1235 "extensions": {
1236 "seed_widgets": {
1237 "10": 0
1238 }
1239 },
1240 "version": 0.4
1241}