Character Reference Pack

Combine separate reference roles into prompts and numbered Qwen image inputs. First Pack: use anchor_prompt to generate the full-body review character. Dataset Pack: connect that approved image to anchor and send character_lock to the shot planner. Prompts/roles come from prompts/character_references.json. This node prepares references; it does not generate images.

By r-vage·Created 11 months ago·Updated about 15 hours ago· 38

face
body
clothing
rear
anchor

character_lock
anchor_prompt
negative
image_1
image_2
image_3
image_4
image_5
report

◄width832►

◄height1216►

◄identityThe same adult character throughout the dataset.►

◄body_description►

◄outfitPlain grey cardigan, blue jeans and white sneakers.►

◄scenePlain white studio background and light floor, soft natural light.►

◄extra_rules►

Category🌒 Eclipse/ Image

Inputs (12)

Name	Type	Default	Description
width	INT	832256–2048	Reference canvas width, 256–2048 in multiples of 32. Images are padded without stretching or cropping. For native Qwen 2.1, use encoder resolution=0 and its latent output to preserve this canvas size.
height	INT	1216256–2048	Reference canvas height, 256–2048 in multiples of 32. Together with width, sets the padded reference dimensions. Share these settings between assembly and dataset Packs for consistent sizing.
identity	STRING	The same adult character throughout the dataset.	Identity details preserved in review and dataset prompts, such as age, hair and distinctive features. Supplements the face image; does not generate a face reference by itself.
body_description	STRING		Describe visible build and relative proportions to preserve, such as chest-to-waist relationship or limb lengths. Supplements a body image, or guides physique without one. Not a measured BMI or numeric anatomy control.
outfit	STRING	Plain grey cardigan, blue jeans and white sneakers.	Fixed wardrobe and explicit choices for missing items. Keep consistent with the clothing reference and its completion inventory; clear the example outfit when using a different reference. Used for assembly and later shots, not for the separate clothing extraction sampler.
scene	STRING	Plain white studio background and light floor, soft natural light.	Background and lighting for anchor_prompt's review image only. Not included in character_lock: set the planner's scene separately for dataset shots, or connect one shared text node to both.
extra_rules	STRING		Additional instructions appended to both character_lock and anchor_prompt. File role/assembly rules still apply; this is not a full override. Avoid fixed poses here if the planner should vary them.
faceopt	IMAGE		Required to execute: one face reference controlling identity, hair and skin tone. Connect the face Select output or a direct image. This role does not control body proportions or wardrobe.
bodyopt	IMAGE		Optional single image controlling visible physique/proportions. Its face and clothing are excluded by the role prompt, but model adherence needs review. Without it, physique comes from the body description/approved character.
clothingopt	IMAGE		Optional outfit reference or completed garment flat lay. Controls garments, materials, colors and accessories, not the original wearer. Check missing pieces before assembly; keep written outfit choices consistent. Underwear stays beneath opaque outer layers when worn.
rearopt	IMAGE		Optional compatible back-view reference for rear hair and garment details when visible. Does not force rear-facing shots or override the selected face, physique or wardrobe.
anchoropt	IMAGE		Approved assembled character image, not a text prompt. Leave empty in the first Pack that creates the review character. Connect the reviewed image here in the dataset Pack to preserve the combined look. Becomes image_1; pose/framing may still vary.

Outputs (9)

Name	Type	Description
character_lock	STRING	Shared identity, physique, wardrobe and numbered reference-role instructions. Connect to Character Shot Planner's character_lock for varied dataset shots. Excludes the fixed full-body review pose and Pack scene.
anchor_prompt	STRING	Complete positive prompt to generate the first full-body REVIEW CHARACTER. Connect to the assembly encoder. Normally unused on the second/dataset Pack, whose character_lock goes to the planner instead. This is text, not the anchor image.
negative	STRING	General assembly/dataset negative from prompts/character_references.json. Connect to the corresponding encoder's negative input. Separate from each Prepare node's role-specific negative override.
image_1	IMAGE	Connect to Qwen encoder image1. Active roles are packed consecutively: anchor if supplied, then face, body, clothing, rear. This socket has no fixed role; check report. Unused slots return no image. Keep paired with this Pack's prompt.
image_2	IMAGE	Connect to Qwen encoder image2. Active roles are packed consecutively: anchor if supplied, then face, body, clothing, rear. This socket has no fixed role; check report. Unused slots return no image. Keep paired with this Pack's prompt.
image_3	IMAGE	Connect to Qwen encoder image3. Active roles are packed consecutively: anchor if supplied, then face, body, clothing, rear. This socket has no fixed role; check report. Unused slots return no image. Keep paired with this Pack's prompt.
image_4	IMAGE	Connect to Qwen encoder image4. Active roles are packed consecutively: anchor if supplied, then face, body, clothing, rear. This socket has no fixed role; check report. Unused slots return no image. Keep paired with this Pack's prompt.
image_5	IMAGE	Connect to Qwen encoder image5. Active roles are packed consecutively: anchor if supplied, then face, body, clothing, rear. This socket has no fixed role; check report. Unused slots return no image. Keep paired with this Pack's prompt.
report	STRING	Lists actual image-number-to-role mapping, canvas size and omitted optional references. Connect to Show Any when checking which image the prompt calls image1, image2, etc.