PradhaanSBhat
Update to gradio 6.30.0
70b3ce2
Raw History Blame Contribute Delete
21.7 kB
"""Thinking in Boxes β€” Hugging Face Spaces (ZeroGPU) demo.
Self-contained Gradio-SDK app for a FREE ZeroGPU Space.
β€’ Generation runs inside @spaces.GPU (the pipeline moves to CUDA per request),
β€’ The 3D box editor is embedded via <iframe srcdoc=...>,
β€’ Place and move boxes by hand.
Space config (README front-matter): sdk: gradio Β· app_file: demo.py
Secret: HF_TOKEN (account that accepted the FLUX.1-Kontext-dev license).
Local run: `python demo.py`.
"""
import base64
import inspect
import io
import json
import os
import random
from pathlib import Path
from types import SimpleNamespace
os.environ.setdefault("HF_ENDPOINT", "https://huggingface.co")
os.environ["GRADIO_TEMP_DIR"] = os.environ.get("GRADIO_TEMP_DIR") or "/tmp/gradio_tib"
os.makedirs(os.environ["GRADIO_TEMP_DIR"], exist_ok=True)
# ZeroGPU: import spaces BEFORE torch β€” it patches torch's CUDA at import time,
# and ZeroGPU detects @spaces.GPU functions through that patch.
try:
import spaces
except Exception:
def _gpu(*a, **k):
if a and callable(a[0]):
return a[0]
def deco(f):
return f
return deco
spaces = SimpleNamespace(GPU=_gpu)
import gradio as gr
import numpy as np
import torch
from diffusers import DiffusionPipeline
from PIL import Image
HERE = Path(__file__).resolve().parent
EDITOR_HTML = HERE / "box-editor.html"
# ── Defaults (mirror inference.sh) ──
DEFAULT_MODEL = "black-forest-labs/FLUX.1-Kontext-dev"
DEFAULT_PIPELINE = "pipelines/pipeline_flux_kontext.py"
DEFAULT_DTYPE = "bfloat16"
DEFAULT_LORA = "pradhaansbhat/Thinking-In-Boxes"
DEFAULT_WEIGHT = "pytorch_lora_weights.safetensors"
DEFAULT_ADAPTER = "100K+20K-LoRA"
DEFAULT_LOCAL = False
DEFAULT_STEPS = 28
DEFAULT_GUIDANCE = 3.5
DEFAULT_TRUE_CFG = 1.0
DEFAULT_SCALE = 1.0
DEFAULT_HEIGHT = 512
DEFAULT_WIDTH = 512
DEFAULT_MAX_SEQ = 512
DTYPE_MAP = {
"bfloat16": torch.bfloat16, "float16": torch.float16, "half": torch.float16,
"float32": torch.float32, "float": torch.float32, "float64": torch.float64, "double": torch.float64,
}
STATE = {"pipe": None, "max_sequence_length": DEFAULT_MAX_SEQ, "last_seed": 0}
# ─────────────────────────────────────────────────────── pipeline (CPU load) ──
def load_pipe_cpu():
"""Load the pipeline on CPU (ZeroGPU exposes CUDA only inside @spaces.GPU)."""
dtype = DTYPE_MAP[DEFAULT_DTYPE]
token = os.environ.get("HF_TOKEN") or None
pipe = DiffusionPipeline.from_pretrained(
DEFAULT_MODEL, custom_pipeline=DEFAULT_PIPELINE or None, trust_remote_code=True,
local_files_only=DEFAULT_LOCAL, token=token, dtype=dtype)
if DEFAULT_LORA:
pipe.transformer.load_lora_adapter(
DEFAULT_LORA, weight_name=DEFAULT_WEIGHT or None, adapter_name=DEFAULT_ADAPTER or None)
print(f"[tib-spaces] LoRA {DEFAULT_ADAPTER} loaded from {DEFAULT_LORA}", flush=True)
pipe.transformer.to(memory_format=torch.channels_last)
pipe.vae.to(memory_format=torch.channels_last)
return pipe
def call_pipe(pipe, **kwargs):
sig = inspect.signature(pipe.__call__)
valid = {k: v for k, v in kwargs.items() if k in sig.parameters}
return pipe(**valid).images
def dataurl_to_pil(u):
if not u:
return None
if "," in u:
u = u.split(",", 1)[1]
return Image.open(io.BytesIO(base64.b64decode(u))).convert("RGB")
# ──────────────────────────────────────────────── generation (ZeroGPU) ──
@spaces.GPU(duration=120)
def generate(image_u, src_u, tgt_u, steps, guidance, true_cfg, lora_scale,
randomize, height, width, num_images, progress=gr.Progress(track_tqdm=True)):
if STATE["pipe"] is None:
raise gr.Error("Model not loaded β€” check the Space logs (HF_TOKEN / license).")
pipe = STATE["pipe"].to("cuda")
src = dataurl_to_pil(src_u); tgt = dataurl_to_pil(tgt_u)
if src is None or tgt is None:
raise gr.Error("Grab the source + target boxes from the editor first.")
image = dataurl_to_pil(image_u)
seed = random.randint(0, 2**31 - 1) if randomize else STATE.get("last_seed", 0)
STATE["last_seed"] = seed
generator = torch.Generator(device="cuda").manual_seed(seed)
images = call_pipe(
pipe,
image=image, source_primitive=src, target_primitive=tgt,
prompt="", negative_prompt=None,
guidance_scale=float(guidance), true_cfg_scale=float(true_cfg),
num_inference_steps=int(steps), num_images_per_prompt=int(num_images),
height=int(height), width=int(width), max_area=int(height) * int(width),
max_sequence_length=int(STATE["max_sequence_length"]), generator=generator,
joint_attention_kwargs={"scale": float(lora_scale)},
)
return list(images), f"seed {seed}"
def grab_show(status, src_u, tgt_u):
msg = (status or "").strip()
if msg.startswith("⚠") or not (src_u and tgt_u):
gr.Warning(msg.lstrip("⚠ ").strip() or "Set a source, edit the boxes, then grab.")
else:
gr.Info("βœ“ Grabbed image + source/target boxes from the editor.")
return dataurl_to_pil(src_u), dataurl_to_pil(tgt_u)
# ── mock detector (no WildDet3D on Spaces): plausible metric boxes, camera frame y-down ──
def run_mock(image_np, prompts, points=None, boxes=None):
H, W = image_np.shape[:2]
fx = 1.2 * W
K = [[fx, 0, W / 2], [0, fx, H / 2], [0, 0, 1]]
prompt_xy = []
if points:
prompt_xy = [(p[0], p[1]) for p in points if int(p[2]) == 1]
elif boxes:
prompt_xy = [((b[0] + b[2]) / 2, (b[1] + b[3]) / 2) for b in boxes]
if prompt_xy:
dets = []
for i, (u, v) in enumerate(prompt_xy):
Z = 6.0
X = (u - W / 2) * Z / fx
Y = (v - H / 2) * Z / fx
dets.append({"label": prompts[i] if i < len(prompts) else f"object {i+1}",
"class_id": i, "score": 0.88,
"box2d": [u - 40, v - 40, u + 40, v + 40],
"center": [X, Y, Z], "dims": [0.8, 1.0, 0.8], "quat": [1, 0, 0, 0]})
return {"image_hw": [H, W], "intrinsics": K, "detections": dets}
catalog = {
"car": {"center": [0.4, 0.9, 8.0], "dims": [1.8, 1.5, 4.3], "quat": [0.966, 0, 0.259, 0]},
"person": {"center": [-2.2, 0.5, 6.0], "dims": [0.6, 1.7, 0.6], "quat": [1, 0, 0, 0]},
"bicycle": {"center": [2.4, 0.8, 5.0], "dims": [0.6, 1.1, 1.7], "quat": [0.985, 0, 0.174, 0]},
"chair": {"center": [0.0, 0.6, 3.2], "dims": [0.6, 0.9, 0.6], "quat": [1, 0, 0, 0]},
"cup": {"center": [0.1, 0.4, 1.4], "dims": [0.09, 0.11, 0.09], "quat": [1, 0, 0, 0]},
}
dets = []
for i, p in enumerate(prompts):
pr = catalog.get(p.strip().lower(), {"center": [0, 0.6, 4 + i], "dims": [0.8, 0.8, 0.8], "quat": [1, 0, 0, 0]})
dets.append({"label": p.strip(), "class_id": i, "score": 0.9 - 0.05 * i,
"box2d": [W * 0.3, H * 0.4, W * 0.7, H * 0.8], **pr})
return {"image_hw": [H, W], "intrinsics": K, "detections": dets}
def detect_objects(payload_json):
try:
req = json.loads(payload_json) if payload_json else {}
except Exception:
return json.dumps({"error": "bad request payload"})
img = dataurl_to_pil(req.get("image"))
if img is None:
return json.dumps({"error": "no image β€” load one in the editor first"})
arr = np.array(img).astype(np.float32)
prompts = [p.strip() for p in (req.get("prompts") or "").split(",") if p.strip()]
points = req.get("points") or None
boxes = req.get("boxes") or None
try:
return json.dumps(run_mock(arr, prompts, points, boxes))
except Exception as e:
return json.dumps({"error": str(e)})
# ───────────────────────────────────────────── editor bridge (browser JS) ──
GRAB_JS = """
() => {
const f = document.getElementById('tib-editor');
if (!f || !f.contentWindow) return ["", "", "", "⚠ Editor not loaded yet."];
let st;
try { st = f.contentWindow.getSourceTarget(); }
catch (e) { return ["", "", "", "⚠ Editor not ready: " + e.message]; }
if (!st || st.error) return ["", "", "", "⚠ " + ((st && st.error) || "no data")];
return [st.image || "", st.source, st.target, "βœ“ Grabbed image + source + target from the editor."];
}
"""
DETECT_BRIDGE_JS = """
() => {
if (window.__tibDetectBridge) return;
window.__tibDetectBridge = true;
const setVal = (el, v) => {
const proto = el.tagName === 'TEXTAREA' ? window.HTMLTextAreaElement.prototype
: window.HTMLInputElement.prototype;
const d = Object.getOwnPropertyDescriptor(proto, 'value');
d && d.set ? d.set.call(el, v) : (el.value = v);
el.dispatchEvent(new Event('input', { bubbles: true }));
};
window.addEventListener('message', (e) => {
const d = e.data; if (!d || d.type !== 'tib-detect') return;
const box = document.querySelector('#tib-detect-payload');
const ta = box && (box.querySelector('textarea') || box.querySelector('input'));
if (!ta) return;
setVal(ta, JSON.stringify(d));
setTimeout(() => { const b = document.querySelector('#tib-detect-trigger');
if (b) (b.querySelector('button') || b).click(); }, 50);
});
}
"""
DETECT_RESULT_JS = """
(resultJson) => {
const f = document.getElementById('tib-editor');
if (!f || !f.contentWindow) return;
let p; try { p = JSON.parse(resultJson || '{}'); } catch (e) { p = { error: 'bad detector result' }; }
const msg = (p && p.error) ? { type: 'tib-detect-result', error: p.error }
: { type: 'tib-detect-result', detections: p };
try { f.contentWindow.postMessage(msg, '*'); } catch (e) {}
}
"""
# ──────────────────────────────────────────────────────────────── theme/CSS ──
TERRACOTTA = gr.themes.colors.Color(
name="terracotta", c50="#faf1ee", c100="#f3d9d0", c200="#e8b5a6", c300="#d98f78",
c400="#cf7458", c500="#c0604a", c600="#a5503d", c700="#7b4f3a", c800="#5f3b2c",
c900="#432a20", c950="#2a1a14")
THEME = gr.themes.Soft(
primary_hue=TERRACOTTA, secondary_hue=TERRACOTTA, neutral_hue=gr.themes.colors.stone,
font=[gr.themes.GoogleFont("Outfit"), "system-ui", "sans-serif"],
font_mono=[gr.themes.GoogleFont("JetBrains Mono"), "monospace"],
)
_VARS = """
--body-background-fill:#F6F3EE; --background-fill-primary:#FFFFFF; --background-fill-secondary:#EDEAE3;
--block-background-fill:#FFFFFF; --block-border-color:#D8D4CB; --border-color-primary:#D8D4CB;
--border-color-accent-subdued:#e6c3ba; --panel-background-fill:#FFFFFF;
--body-text-color:#1A1714; --body-text-color-subdued:#6B6560;
--block-label-text-color:#C0604A; --block-label-background-fill:#F1EBE3; --block-title-text-color:#C0604A;
--button-primary-background-fill:#C0604A; --button-primary-background-fill-hover:#7B4F3A;
--button-primary-text-color:#FFFFFF; --button-primary-border-color:#C0604A;
--button-secondary-background-fill:#FFFFFF; --button-secondary-background-fill-hover:#EDEAE3;
--button-secondary-border-color:#D8D4CB; --button-secondary-text-color:#1A1714;
--input-background-fill:#FFFFFF; --input-border-color:#D8D4CB; --neutral-800:#3a352f;
--color-accent:#C0604A; --color-accent-soft:#f3d9d0; --link-text-color:#C0604A;
--checkbox-background-color:#FFFFFF; --checkbox-border-color:#C0604A;
--checkbox-background-color-selected:#C0604A; --checkbox-border-color-selected:#C0604A;
--checkbox-background-color-focus:#C0604A; --checkbox-border-color-focus:#C0604A;
"""
CSS = """
@import url('https://fonts.googleapis.com/css2?family=Cormorant+Garamond:ital,wght@0,500;0,600;1,500;1,600&display=swap');
:root, .gradio-container, gradio-app, .dark { __VARS__ }
body, .gradio-container, gradio-app, .dark { background:#F6F3EE !important; color:#1A1714 !important; }
.gradio-container{max-width:100%!important;width:100%!important;margin:0!important;padding:0 28px 28px!important}
.gradio-container .main, .main.fillable, .fillable.app, main.contain{max-width:100%!important}
.dark input, .dark textarea{background:#FFFFFF!important;color:#1A1714!important}
.gradio-container .block > label > span:first-child,
.gradio-container .block .head > label > span:first-child,
.gradio-container .block > .container > span:first-child,
.gradio-container .block .wrap > .head span{color:#C0604A!important;background:#F1EBE3!important}
.gradio-container label > span{color:#C0604A}
button.primary, .primary{color:#FFF6F1!important}
.gr-button-primary, button[variant="primary"]{color:#FFF6F1!important}
#tib-header{position:relative;text-align:center;margin:2px 0 14px;padding:22px 0 18px;overflow:hidden}
#tib-header .v{position:relative;z-index:2;display:inline-block;font-family:'Outfit',system-ui,sans-serif;font-weight:600;font-size:.78rem;letter-spacing:.1em;text-transform:uppercase;color:#C0604A;background:#F1EBE3;border:1px solid #e6c3ba;border-radius:999px;padding:4px 15px;margin-bottom:12px}
#tib-header .t{position:relative;z-index:2;font-family:'Cormorant Garamond',Georgia,serif;font-weight:600;font-size:3.1rem;line-height:1;letter-spacing:-.01em;color:#1A1714}
#tib-header .t em{font-style:italic;color:#C0604A}
#tib-header .s{position:relative;z-index:2;font-family:'Outfit',sans-serif;font-weight:400;color:#6B6560;font-size:1.06rem;margin-top:6px;letter-spacing:.01em}
.tib-cube{position:absolute;z-index:1;opacity:.5;animation:tibfloat 9s ease-in-out infinite}
.tib-cube.c1{left:6%;top:24%;width:54px;animation-delay:0s}
.tib-cube.c2{left:16%;top:58%;width:38px;opacity:.35;animation-delay:1.4s}
.tib-cube.c3{right:8%;top:20%;width:60px;animation-delay:.6s}
.tib-cube.c4{right:19%;top:60%;width:40px;opacity:.35;animation-delay:2.1s}
.tib-cube.c5{left:38%;top:8%;width:30px;opacity:.28;animation-delay:1.1s}
.tib-cube.c6{right:36%;top:70%;width:32px;opacity:.3;animation-delay:.3s}
@keyframes tibfloat{0%,100%{transform:translateY(0) rotate(-6deg)}50%{transform:translateY(-12px) rotate(6deg)}}
h2,h3{font-family:'Cormorant Garamond',Georgia,serif!important;color:#1A1714}
.tib-edcol{display:flex!important;flex-direction:column;min-height:0}
.tib-edcol > *, .tib-edcol .block, .tib-edcol .html-container, .tib-edcol .prose{
display:flex!important;flex-direction:column;flex:1 1 auto!important;min-height:0;height:auto!important}
#tib-editor{flex:1 1 auto;height:100%!important;min-height:620px;display:block}
.tib-ctrl{align-self:stretch;flex-wrap:nowrap!important;overflow-y:auto;padding-right:8px;gap:12px}
.tib-ctrl .prose:empty, .tib-ctrl .md:empty, .tib-ctrl .block.padded:empty{display:none!important;min-height:0!important;padding:0!important;margin:0!important}
.tib-sq .image-container{aspect-ratio:1/1;min-height:190px}
.tib-sq .image-container img{object-fit:contain}
.tib-out .wrap, .tib-out .empty, .tib-out .grid-wrap, .tib-out .fixed-height, .tib-out .preview{
aspect-ratio:1/1!important;height:auto!important;min-height:0!important;max-height:none!important;
display:flex!important;align-items:center!important;justify-content:center!important;overflow:hidden}
.tib-out .grid-container{width:100%!important;height:100%!important;display:grid!important;grid-template-columns:1fr!important;place-items:center!important}
.tib-out .thumbnail-item, .tib-out .image-container{width:100%!important;height:100%!important;display:flex!important;align-items:center!important;justify-content:center!important}
.tib-out img{max-width:100%!important;max-height:100%!important;width:auto!important;height:auto!important;object-fit:contain!important;margin:auto!important}
/* accordion titles + section headings readable on cream (were invisible) */
.gradio-container .label-wrap, .gradio-container .label-wrap > span, .gradio-container .label-wrap span{color:#1A1714!important}
.gradio-container h4, .gradio-container .prose h4{color:#1A1714!important}
.tib-step button, button.tib-step{position:relative;padding-left:46px!important;font-weight:600!important}
.tib-step button::before, button.tib-step::before{
position:absolute;left:12px;top:50%;transform:translateY(-50%);
width:24px;height:24px;border-radius:50%;display:flex;align-items:center;justify-content:center;
font-family:'Outfit',system-ui,sans-serif;font-weight:700;font-size:.88rem;line-height:1;border:1.5px solid #C0604A}
.tib-step1 button::before, button.tib-step1::before{content:"1";background:#C0604A;color:#FFF6F1}
.tib-step2 button::before, button.tib-step2::before{content:"2";background:#FFF6F1;color:#C0604A;border-color:#FFF6F1}
.gradio-container input[type=checkbox]{accent-color:#C0604A;width:18px;height:18px;border:1.5px solid #C0604A!important;background:#FFFFFF!important;appearance:auto!important;-webkit-appearance:auto!important;opacity:1!important}
.gradio-container input[type=checkbox]:checked{background:#C0604A!important;border-color:#C0604A!important}
.tib-hidden{display:none!important}
footer{display:none!important}
""".replace("__VARS__", _VARS)
def _editor_srcdoc():
"""Inline box-editor.html into an iframe srcdoc (Gradio SDK serves no custom route)."""
html = EDITOR_HTML.read_text(encoding="utf-8")
esc = html.replace("&", "&amp;").replace('"', "&quot;")
return ('<iframe id="tib-editor" srcdoc="' + esc + '" '
'style="width:100%;height:100%;min-height:620px;border:1px solid #D8D4CB;'
'border-radius:12px;background:#F6F3EE"></iframe>')
def build_demo():
iframe = _editor_srcdoc()
with gr.Blocks(title="Thinking in Boxes") as demo:
cube = ('<svg class="tib-cube {c}" viewBox="0 0 100 100" fill="none" stroke="#C0604A" '
'stroke-width="4" stroke-linejoin="round">'
'<path d="M20 35 L50 20 L80 35 L50 50 Z"/>'
'<path d="M20 35 L20 70 L50 85 L50 50 Z"/>'
'<path d="M80 35 L80 70 L50 85 L50 50 Z"/></svg>')
cubes = "".join(cube.format(c=f"c{i}") for i in range(1, 7))
gr.HTML(f'<div id="tib-header">{cubes}'
'<div class="v">NeurIPS 2026</div>'
'<div class="t">Thinking in <em>Boxes</em></div>'
'<div class="s">3D Editing in Real Images Made Easy</div></div>')
with gr.Row(equal_height=True):
with gr.Column(scale=3, elem_classes="tib-edcol"):
gr.HTML(iframe)
with gr.Column(scale=2, elem_classes="tib-ctrl"):
grab = gr.Button("Grab image + boxes from editor", elem_classes="tib-step tib-step1")
with gr.Row(equal_height=True):
src_prev = gr.Image(label="Source boxes", elem_classes="tib-sq", show_label=True)
tgt_prev = gr.Image(label="Target boxes", elem_classes="tib-sq", show_label=True)
image_u = gr.Textbox(elem_classes="tib-hidden"); src_u = gr.Textbox(elem_classes="tib-hidden")
tgt_u = gr.Textbox(elem_classes="tib-hidden"); status_u = gr.Textbox(elem_classes="tib-hidden")
randomize = gr.Checkbox(value=True, label="🎲 Randomize seed")
gen = gr.Button("Generate", variant="primary", elem_classes="tib-step tib-step2")
out = gr.Gallery(label="Generated edit", columns=1, object_fit="contain",
show_label=True, elem_classes="tib-out")
seed_used = gr.Markdown()
detect_payload = gr.Textbox(elem_id="tib-detect-payload", elem_classes="tib-hidden")
detect_result = gr.Textbox(elem_id="tib-detect-result", elem_classes="tib-hidden")
detect_trigger = gr.Button("detect", elem_id="tib-detect-trigger", elem_classes="tib-hidden")
with gr.Accordion("Generation settings", open=False):
with gr.Row():
steps = gr.Slider(1, 100, value=DEFAULT_STEPS, step=1, label="Inference steps")
num_images = gr.Slider(1, 4, value=1, step=1, label="Images per run")
with gr.Row():
height = gr.Slider(256, 1536, value=DEFAULT_HEIGHT, step=16, label="Height")
width = gr.Slider(256, 1536, value=DEFAULT_WIDTH, step=16, label="Width")
guidance = gr.Number(value=DEFAULT_GUIDANCE, elem_classes="tib-hidden")
true_cfg = gr.Number(value=DEFAULT_TRUE_CFG, elem_classes="tib-hidden")
lora_scale = gr.Number(value=DEFAULT_SCALE, elem_classes="tib-hidden")
grab.click(None, None, [image_u, src_u, tgt_u, status_u], js=GRAB_JS).then(
grab_show, [status_u, src_u, tgt_u], [src_prev, tgt_prev])
detect_trigger.click(detect_objects, [detect_payload], [detect_result]).then(
None, [detect_result], None, js=DETECT_RESULT_JS)
demo.load(None, None, None, js=DETECT_BRIDGE_JS)
gen.click(generate,
[image_u, src_u, tgt_u, steps, guidance, true_cfg, lora_scale,
randomize, height, width, num_images],
[out, seed_used])
return demo
try:
STATE["pipe"] = load_pipe_cpu()
print("[tib-spaces] pipeline ready on CPU (moves to GPU per request).", flush=True)
except Exception:
import traceback; traceback.print_exc()
demo = build_demo()
if __name__ == "__main__":
demo.launch(theme=THEME, css=CSS, ssr_mode=False,
favicon_path=str(HERE / "assets" / "favicon.svg"),share=True)