ComfyUI Node
⚗ Alchemica Hybrid (fuse the traits of photos)
Fuses the traits of up to 4 photos into ONE new subject described in words. The text encoder reads each photo (the face from a close crop), code shares the traits by weight (whole or blended), the text encoder writes the final description.
⚗ Alchemica Hybrid (fuse the traits of photos)
- clip
- image_1
- image_2
- image_3
- image_4
- subject
- report
◄weight_11.00►
◄weight_21.00►
◄weight_31.00►
◄weight_41.00►
◄idea►
◄subjectcreature / animal►
◄words70►
◄creativity0.00►
◄seed0►
◄mix_modeone trait = one photo►
◄idea_prioritybalanced►
◄unique_facetrue►
◄face_croptrue►
Category⚗ AlchemicaMente/characters
Inputs (18)
| Name | Type | Default | Description |
|---|---|---|---|
| clip | CLIP | Krea 2's text encoder (Qwen3-VL-4B): it looks at the photos and writes the description | |
| image_1 | IMAGE | — | |
| weight_1 | FLOAT | 1.000–4 | Weight of photo 1. Normalised: it sets the SHARE of this photo (1 and 1 = half each; 2 and 1 = two thirds from photo 1). 0 = ignore |
| weight_2 | FLOAT | 1.000–4 | Weight of photo 2. Normalised: it sets the SHARE of this photo (1 and 1 = half each; 2 and 1 = two thirds from photo 1). 0 = ignore |
| weight_3 | FLOAT | 1.000–4 | Weight of photo 3. Normalised: it sets the SHARE of this photo (1 and 1 = half each; 2 and 1 = two thirds from photo 1). 0 = ignore |
| weight_4 | FLOAT | 1.000–4 | Weight of photo 4. Normalised: it sets the SHARE of this photo (1 and 1 = half each; 2 and 1 = two thirds from photo 1). 0 = ignore |
| idea | STRING | Optional: extra instruction for the final description. How much it weighs against the photos: 'priorita_idea' | |
| subject | COMBO | creature / animal | Which traits are compared: animal (body, coat, ears, muzzle, eyes, tail...), person (17 traits, 14 of the face), object, or generic |
| words | INT | 7030–220 | Length of the hybrid description. Person: 110-140 (17 traits + the idea) |
| creativity | FLOAT | 0.000–1.5 | 0 = deterministic (same photos + weights = same description: use it for datasets). 0.5-0.9 = wording variants, change the seed. The allocation of traits never changes |
| seed | INT | 00–18446744073709550000 | Only matters when creativity > 0 |
| image_2opt | IMAGE | — | |
| image_3opt | IMAGE | — | |
| image_4opt | IMAGE | — | |
| mix_modeopt | COMBO | one trait = one photo | one trait = one photo: every trait comes whole from one photo (weights = how MANY traits). blended traits: every trait is an in-between of all photos (weights = how MUCH each pulls): the real mix of the characters |
| idea_priorityopt | COMBO | balanced | How much the idea weighs against the photos. 'traits win': the idea sets pose, clothes and mood but never changes the traits from the photos (person: the FACE and hair stay from the photos, the idea may still set the body). 'idea wins': the idea overrides them |
| unique_faceopt | BOOLEAN | true | Person: removes generic beauty words (perfect, flawless, beautiful, symmetrical...) that pull every face to the model's average face, and asks for a specific individual |
| face_cropopt | BOOLEAN | true | Person: reads the face from a close crop (insightface detector, CPU). On a full-body photo the face is too small to describe. Off = whole photo, as in 1.0.x |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| subject | STRING | — |
| report | STRING | — |