Ju_HumanSD_A_Native_ICCV_2023_supplemental - Image Style and Conditioning Comparison

Aug 28, 2026

Supplemental material for Ju_HumanSD_A_Native_ICCV_2023 focuses on image generation under varying conditioning signals and image style settings. The content organizes multiple prompts describing humans, scenes, actions, and environments, then compares outputs produced by several methods including SD, Pose, ControlNet, Adapter, and Ours. It also lists style-related descriptors (e.g., long shot, HDR, lighting and emotion cues) and provides selected versus non-selected image outcomes. The examples cover both single and composite figures, with diverse subjects and settings.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Image styles and descriptors

Conditioning methods comparison

Example prompts and generated scenes

Strategy Example rough description < adj. > image style (14 styles) < n. > preposition < prep. > an extremely detailed a realistic … photo sculpture … depicting of … | human number
< num. >
human shape
< adj. >
human age and sex
< adj. > + < n. >
action (from 5 datasets)
< v. > | | one single
two separate
…
thin
strong
…
young girl
old man
…
run
climb up a ladder
… | | --- | --- | --- | | scene
< n. >
time
< adv. >
weather
< adv. >
human position
< adv. > | in the street of London
in a library with piles of books …
at midnight
in the morning …
in a rainy day
in warm spring …
facing left
facing forward … | | image style emphasize < n. > by Norman Rockwell a close up statue … camera composition < n. > view < n. > light < n. > emotion < n. > magic words < n. > long shot wide shot … ahead view tilted … fluorescent lighting strong shadows … threatening spirited … masterpiece HDR … | (an extremely detailed) (photo) (depicting) (one single) | | --- | |
(long shot),(threatening),(masterpiece),(HDR) | | Selected Not Selected Image Human Scene | A girl is running in the field A man standing in front of the White House (a) SD (b) Pose (c) ControlNet (d) Adapter (e) Ours (a) SD (b) Pose (c) ControlNet (d) Adapter (e) Ours A man looking at a big mountain (a) SD (b) Pose (c) ControlNet (d) Adapter (e) Ours (a) SD (b) Pose (c) ControlNet (d) Adapter (e) Ours A sketch depicting the Iron Man with hair (a) SD (b) Pose (c) ControlNet (d) Adapter (e) Ours A sketch depicting a man in the office (a) SD (b) Pose (c) ControlNet (d) Adapter (e) Ours A Chinese shadow play ofa woman standing (a) SD (b) Pose (c) ControlNet (d) Adapter (e) Ours A Kids Drawing depicting a girl (a) SD (b) Pose (c) ControlNet (d) Adapter (e) Ours A extremely detailed oilpainting ofa beautiful woman face (a) SD (b) Pose (c) ControlNet (d) Adapter (e) Ours apiece of digital art ofa superwoman (a) SD (b) Pose (c) ControlNet (d) Adapter (e) Ours (a) (b) (c) Cartoon, a cartoon girl jumping with hearts in the air Cartoon, a scarecrow in the woods Mural, an oldpainting ofan angel Mural, an oldpainting ofan angel Sketch, man doing various poses Garage kits, a stuffedanimalwith a smalltoy Relief, aplaque with a man holding a flag anda boy Oilpainting, a man anda woman dancing ballet Cosplay, a woman with swordin her hand (a) Pose (c) ControlNet (d) Ours Digitalart, two people standing in the milky sky Children painting, a family with sun in notebook Mural, a mosaic of a man anda woman dancing Shadow play, a set of decorative items with people Sketch, a gymnast doing a handstand Garage kits, a girl laying on the ground Watercolor, a cucoloris looking at sky Oilpainting, girls dancing in a theater Chinese painting, a beautiful woman in Peking Opera (a) Pose (c) ControlNet (d) Ours Chinese inkpainting, a woman sit in front ofmontains Kids drawing, man kneel down to woman with flowers Wallpainting, boys anda dog pulling a cart Shadow play, a couple in a room Garage kits, a girl with pinkshirt orange hat Garage kits, a girl in blackandredoutfit Ukiyoe, two man talking with each other Sculpture, a strong man sitting on a rock Oilpainting, the queen Play the harp in the forest (a) Pose (c) ControlNet (d) Ours

FAQs

This supplemental material compares which generation methods?

It compares SD, Pose, ControlNet, Adapter, and Ours using the same or related prompts to show differences in the generated images.

What kinds of prompts are used in the examples?

The document uses prompts describing human identity attributes (age, sex, shape), actions (running, climbing), and scenes (street locations, libraries, time and weather) to guide generation.

How are image styles represented in the examples?

It includes an image style selection such as emphasize, long shot, HDR, camera composition, and lighting/emotion-related descriptors to influence the final look.

Aug 28, 2026

Uploaded by Genevieve