oobabooga / text-generation-webui

A Gradio web UI for Large Language Models.
GNU Affero General Public License v3.0
39.44k stars 5.19k forks source link

sd_api_pictures - Default Mode Bug #2943

Closed altoiddealer closed 1 year ago

altoiddealer commented 1 year ago

Describe the bug

Using settings.yaml to set the default mode to "2" (Picture/Adventure mode) does not work correctly.

The UI does show Picture/Adventure mode as being set, but it will not send picture replies until the UI menu is toggled.

NOTE: I tried changing the default "Chat/Chat Type" settings for my model to see if that had any bearing on it (would default to "Instruct" otherwise) but this does not seem to be part of the problem

I've been playing around with the code and I don't know how to fix it (I barely know how to code).

Is there an existing issue for this?

Reproduction

-add to settings.yaml: sd_api_pictures-mode: 2 -Launch the webui. Mode will show as Picture/Adventure mode. -Generate something. Response will be text without image.

Screenshot

Mode

Logs

bin C:\0_SD\oobabooga\installer_files\env\lib\site-packages\bitsandbytes\libbitsandbytes_cuda117.dll
2023-06-30 12:44:29 INFO:Loading settings from settings.yaml...
2023-06-30 12:44:29 INFO:Loading the extension "gallery"...
2023-06-30 12:44:29 INFO:Loading the extension "sd_api_pictures"...
Running on local URL:  http://127.0.0.1:7862

To create a public link, set `share=True` in `launch()`.
2023-06-30 12:44:42 INFO:Loading airoboros-13b-gpt4-1.4-GPTQ-4bit-128g...
2023-06-30 12:44:48 INFO:Loaded the model in 5.84 seconds.

You: Sindel from Mortal Kombat
MintierSD: hyper-realistic wide shot erotic photo, nude Sindel from Mortal Kombat, table dance bar with many guests, sweaty skin, realistic face, large breasts, slim waist, big ass, long legs, gothic jewelry, erotic dance, intricate, cinematic light, high saturation, high contrast, hdr, Hasselblad H6D-400c
You: lustful bartender
MintierSD: realistic photo, wild look, tattooed lustful Finnish female bartender, deep red lips, piercing green eyes, pouring glass of beer, one hand on hip, exposing tight black corset, white apron, wet skin, lace bra, sensuous smile, dimly lit pub, warm ambient atmosphere, moist skin, refined lighting, bokeh effect, beer foam, mirror behind counter, wooden paneling, well lit, DSLR camera, shallow depth of field, fast prime lens, high resolution
You: black woman, coming out of a painting
MintierSD: full length portrait, sexy cartoon Sierra Leonean girl, emerging from an art canvas, surreal, dreamy, whimsical, magical, colorful, vibrant, abstract, fantasy, impressionist, oil painting, thick brush strokes, impasto, high texture, dynamic composition, loose brushwork, vibrant colors, bold shapes, stylized figures, atmospheric lighting, dramatic lighting, strong highlights, rich shadows, textured surfaces, glossy finish, smooth surface, matte finish, metallic sheen, reflective surfaces, dimensionality, depth, three-dimensional quality, illusion of depth, movement, energy, vitality, dynamism, action, motion, excitement, wonder, magic, mystery, enchantment, charm, allure
You: cave explorer
MintierSD: realistic documentary photograph, scantily clad Norfolk Islander woman, wearing headlamp helmet, dirty skin, huge breasts exposed, exploring underground cavern, crawling in narrow tunnel, low ceiling, uneven ground, rough terrain, khaki shorts, muddy boots, eager expression, glossy eyes, determined, reflective surfaces, rocks, dirt, moss, fungi, dark corners, mysterious atmosphere, spelunking equipment, ropes, climbing gear, harness, backpack, dust, darkness, stalactites, stalagmites, Canon EOS 1D X Mark II
You: cartoon Dove Cameron
MintierSD: full body illustration, animated Dove Cameron, voluptuous breasts, round buttocks, cheeky grin, holding microphone, singing, performing, stage presence, pop culture icon, playful, exaggerated features, caricatured, simplified design, minimal background, flat color scheme, bold lines, simple shape language, graphic style, clean line work, digital art, vector graphics
You: Egyptian thief
MintierSD: realistic full body photo, high angle shot, Egyptian thief woman, half naked, seductive pose, ancient pyramid vault, sitting on pile of gold coins, golden goblet in hand, wearing extravagant jewelry, pharaoh headdress, treasure scattered everywhere, stone walls inscribed with hieroglyphs, massive cleavage, curly hair, jewels, precious gems, rubies, diamonds, emeralds, gold bars, shiny golden objects, candlelight, wicked smirk, glittering eyes, lip ring, nose ring, sultry, devious, intricate, slow shutter speed, wide aperture, Sony Alpha a7S III
You: elegant bartender
MintierSD:
--------------------

Output generated in 3.92 seconds (25.50 tokens/s, 100 tokens, context 883, seed 1279675174)

System Info

Windows 11.  RTX 4070ti
altoiddealer commented 1 year ago

I reported two issues initially - the first, having to do with "Seed Resize" defaulting to -1x-1 via the A1111 API.

It turns out, the issue I was experiencing was not due to this, and a few luck-based image generations cemented this as the culprit... but further testing proved otherwise.

github-actions[bot] commented 1 year ago

This issue has been closed due to inactivity for 30 days. If you believe it is still relevant, please leave a comment below.