lovis93 commited on
Commit
8e8f350
·
verified ·
1 Parent(s): 849fa76

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +426 -0
README.md ADDED
@@ -0,0 +1,426 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: Qwen/Qwen-Image-Edit-2511
4
+ tags:
5
+ - qwen
6
+ - qwen-image-edit
7
+ - qwen-image-edit-2511
8
+ - lora
9
+ - multi-angle
10
+ - camera-angles
11
+ - camera-control
12
+ - image-editing
13
+ - image-to-image
14
+ - gaussian-splatting
15
+ - diffusers
16
+ - fal
17
+ language:
18
+ - en
19
+ pipeline_tag: image-to-image
20
+ library_name: diffusers
21
+ ---
22
+
23
+ # Qwen-Image-Edit-2511-Multiple-Angles-LoRA
24
+
25
+ > **The first multi-angle camera control LoRA for Qwen-Image-Edit-2511**
26
+ >
27
+ > 96 camera positions • Trained on 3000+ Gaussian Splatting renders • Built with [fal.ai](https://fal.ai)
28
+
29
+ ---
30
+
31
+ ## Highlights
32
+
33
+ | Feature | Details |
34
+ |---------|---------|
35
+ | **96 Camera Poses** | 4 elevations × 8 azimuths × 3 distances |
36
+ | **3000+ Training Pairs** | Massive dataset for maximum precision |
37
+ | **Gaussian Splatting Data** | High-quality 3D-consistent renders |
38
+ | **Low-Angle Support** | Proper low-angle (-30°) camera control |
39
+ | **Extensively Tested** | More iterations and quality checks |
40
+
41
+ ---
42
+
43
+ ## Results
44
+
45
+ ![Camera Animation Results](all_animations_combined.gif)
46
+
47
+ ---
48
+
49
+ ## Camera System Diagrams
50
+
51
+ ![All 96 Poses](poses_96_animated.gif)
52
+
53
+ ![Distance Comparison](poses_96_distance_comparison.png)
54
+
55
+ ---
56
+
57
+ ## Why This LoRA?
58
+
59
+ **This is the first multi-angle camera control LoRA for [Qwen-Image-Edit-2511](https://huggingface.co/Qwen/Qwen-Image-Edit-2511).**
60
+
61
+ While Qwen-Image-Edit-2511 has built-in viewpoint capabilities, this LoRA provides:
62
+
63
+ - **96 precise camera poses** - Exact control over camera position
64
+ - **3000+ training pairs** - Massive dataset for maximum accuracy
65
+ - **Gaussian Splatting data** - 3D-consistent training for better spatial understanding
66
+ - **Low-angle excellence** - Proper support for ground-level and low camera positions (-30°)
67
+
68
+ ---
69
+
70
+ ## Prompt Format
71
+
72
+ ```
73
+ <sks> [azimuth] [elevation] [distance]
74
+ ```
75
+
76
+ ### Quick Examples
77
+
78
+ ```
79
+ <sks> front view eye-level shot medium shot
80
+ <sks> right side view high-angle shot close-up
81
+ <sks> back view low-angle shot wide shot
82
+ <sks> front-left quarter view elevated shot medium shot
83
+ ```
84
+
85
+ ---
86
+
87
+ ## 96 Camera Positions
88
+
89
+ **4 Elevations × 8 Azimuths × 3 Distances = 96 Poses**
90
+
91
+ ### Azimuths (Horizontal Rotation)
92
+
93
+ ```
94
+
95
+ (front view)
96
+
97
+ 315° │ 45°
98
+ (front-left) │ (front-right)
99
+ ╲ │ ╱
100
+ ╲ │ ╱
101
+ ╲ │ ╱
102
+ 270° ─────────────── ● ─────────────── 90°
103
+ (left side) OBJECT (right side)
104
+ ╱ │ ╲
105
+ ╱ │ ╲
106
+ ╱ │ ╲
107
+ 225° │ 135°
108
+ (back-left) │ (back-right)
109
+
110
+ 180°
111
+ (back view)
112
+ ```
113
+
114
+ | Angle | Descriptor |
115
+ |-------|------------|
116
+ | 0° | `front view` |
117
+ | 45° | `front-right quarter view` |
118
+ | 90° | `right side view` |
119
+ | 135° | `back-right quarter view` |
120
+ | 180° | `back view` |
121
+ | 225° | `back-left quarter view` |
122
+ | 270° | `left side view` |
123
+ | 315° | `front-left quarter view` |
124
+
125
+ ### Elevations (Vertical Angle)
126
+
127
+ | Angle | Descriptor | Description |
128
+ |-------|------------|-------------|
129
+ | -30° | `low-angle shot` | Camera below, looking up |
130
+ | 0° | `eye-level shot` | Camera at object level |
131
+ | 30° | `elevated shot` | Camera slightly above |
132
+ | 60° | `high-angle shot` | Camera high, looking down |
133
+
134
+ ### Distances
135
+
136
+ | Factor | Descriptor | Usage |
137
+ |--------|------------|-------|
138
+ | ×0.6 | `close-up` | Details, textures |
139
+ | ×1.0 | `medium shot` | Balanced, standard |
140
+ | ×1.8 | `wide shot` | Context, environment |
141
+
142
+ ---
143
+
144
+ ## All 96 Prompts Reference
145
+
146
+ ### CLOSE-UP (32 prompts)
147
+
148
+ <details>
149
+ <summary>Click to expand</summary>
150
+
151
+ **Low-angle (-30°)**
152
+ ```
153
+ <sks> front view low-angle shot close-up
154
+ <sks> front-right quarter view low-angle shot close-up
155
+ <sks> right side view low-angle shot close-up
156
+ <sks> back-right quarter view low-angle shot close-up
157
+ <sks> back view low-angle shot close-up
158
+ <sks> back-left quarter view low-angle shot close-up
159
+ <sks> left side view low-angle shot close-up
160
+ <sks> front-left quarter view low-angle shot close-up
161
+ ```
162
+
163
+ **Eye-level (0°)**
164
+ ```
165
+ <sks> front view eye-level shot close-up
166
+ <sks> front-right quarter view eye-level shot close-up
167
+ <sks> right side view eye-level shot close-up
168
+ <sks> back-right quarter view eye-level shot close-up
169
+ <sks> back view eye-level shot close-up
170
+ <sks> back-left quarter view eye-level shot close-up
171
+ <sks> left side view eye-level shot close-up
172
+ <sks> front-left quarter view eye-level shot close-up
173
+ ```
174
+
175
+ **Elevated (30°)**
176
+ ```
177
+ <sks> front view elevated shot close-up
178
+ <sks> front-right quarter view elevated shot close-up
179
+ <sks> right side view elevated shot close-up
180
+ <sks> back-right quarter view elevated shot close-up
181
+ <sks> back view elevated shot close-up
182
+ <sks> back-left quarter view elevated shot close-up
183
+ <sks> left side view elevated shot close-up
184
+ <sks> front-left quarter view elevated shot close-up
185
+ ```
186
+
187
+ **High-angle (60°)**
188
+ ```
189
+ <sks> front view high-angle shot close-up
190
+ <sks> front-right quarter view high-angle shot close-up
191
+ <sks> right side view high-angle shot close-up
192
+ <sks> back-right quarter view high-angle shot close-up
193
+ <sks> back view high-angle shot close-up
194
+ <sks> back-left quarter view high-angle shot close-up
195
+ <sks> left side view high-angle shot close-up
196
+ <sks> front-left quarter view high-angle shot close-up
197
+ ```
198
+
199
+ </details>
200
+
201
+ ### MEDIUM SHOT (32 prompts)
202
+
203
+ <details>
204
+ <summary>Click to expand</summary>
205
+
206
+ **Low-angle (-30°)**
207
+ ```
208
+ <sks> front view low-angle shot medium shot
209
+ <sks> front-right quarter view low-angle shot medium shot
210
+ <sks> right side view low-angle shot medium shot
211
+ <sks> back-right quarter view low-angle shot medium shot
212
+ <sks> back view low-angle shot medium shot
213
+ <sks> back-left quarter view low-angle shot medium shot
214
+ <sks> left side view low-angle shot medium shot
215
+ <sks> front-left quarter view low-angle shot medium shot
216
+ ```
217
+
218
+ **Eye-level (0°)** — Reference pose: `front view eye-level shot medium shot`
219
+ ```
220
+ <sks> front view eye-level shot medium shot
221
+ <sks> front-right quarter view eye-level shot medium shot
222
+ <sks> right side view eye-level shot medium shot
223
+ <sks> back-right quarter view eye-level shot medium shot
224
+ <sks> back view eye-level shot medium shot
225
+ <sks> back-left quarter view eye-level shot medium shot
226
+ <sks> left side view eye-level shot medium shot
227
+ <sks> front-left quarter view eye-level shot medium shot
228
+ ```
229
+
230
+ **Elevated (30°)**
231
+ ```
232
+ <sks> front view elevated shot medium shot
233
+ <sks> front-right quarter view elevated shot medium shot
234
+ <sks> right side view elevated shot medium shot
235
+ <sks> back-right quarter view elevated shot medium shot
236
+ <sks> back view elevated shot medium shot
237
+ <sks> back-left quarter view elevated shot medium shot
238
+ <sks> left side view elevated shot medium shot
239
+ <sks> front-left quarter view elevated shot medium shot
240
+ ```
241
+
242
+ **High-angle (60°)**
243
+ ```
244
+ <sks> front view high-angle shot medium shot
245
+ <sks> front-right quarter view high-angle shot medium shot
246
+ <sks> right side view high-angle shot medium shot
247
+ <sks> back-right quarter view high-angle shot medium shot
248
+ <sks> back view high-angle shot medium shot
249
+ <sks> back-left quarter view high-angle shot medium shot
250
+ <sks> left side view high-angle shot medium shot
251
+ <sks> front-left quarter view high-angle shot medium shot
252
+ ```
253
+
254
+ </details>
255
+
256
+ ### WIDE SHOT (32 prompts)
257
+
258
+ <details>
259
+ <summary>Click to expand</summary>
260
+
261
+ **Low-angle (-30°)**
262
+ ```
263
+ <sks> front view low-angle shot wide shot
264
+ <sks> front-right quarter view low-angle shot wide shot
265
+ <sks> right side view low-angle shot wide shot
266
+ <sks> back-right quarter view low-angle shot wide shot
267
+ <sks> back view low-angle shot wide shot
268
+ <sks> back-left quarter view low-angle shot wide shot
269
+ <sks> left side view low-angle shot wide shot
270
+ <sks> front-left quarter view low-angle shot wide shot
271
+ ```
272
+
273
+ **Eye-level (0°)**
274
+ ```
275
+ <sks> front view eye-level shot wide shot
276
+ <sks> front-right quarter view eye-level shot wide shot
277
+ <sks> right side view eye-level shot wide shot
278
+ <sks> back-right quarter view eye-level shot wide shot
279
+ <sks> back view eye-level shot wide shot
280
+ <sks> back-left quarter view eye-level shot wide shot
281
+ <sks> left side view eye-level shot wide shot
282
+ <sks> front-left quarter view eye-level shot wide shot
283
+ ```
284
+
285
+ **Elevated (30°)**
286
+ ```
287
+ <sks> front view elevated shot wide shot
288
+ <sks> front-right quarter view elevated shot wide shot
289
+ <sks> right side view elevated shot wide shot
290
+ <sks> back-right quarter view elevated shot wide shot
291
+ <sks> back view elevated shot wide shot
292
+ <sks> back-left quarter view elevated shot wide shot
293
+ <sks> left side view elevated shot wide shot
294
+ <sks> front-left quarter view elevated shot wide shot
295
+ ```
296
+
297
+ **High-angle (60°)**
298
+ ```
299
+ <sks> front view high-angle shot wide shot
300
+ <sks> front-right quarter view high-angle shot wide shot
301
+ <sks> right side view high-angle shot wide shot
302
+ <sks> back-right quarter view high-angle shot wide shot
303
+ <sks> back view high-angle shot wide shot
304
+ <sks> back-left quarter view high-angle shot wide shot
305
+ <sks> left side view high-angle shot wide shot
306
+ <sks> front-left quarter view high-angle shot wide shot
307
+ ```
308
+
309
+ </details>
310
+
311
+ ---
312
+
313
+ ## Files
314
+
315
+ | File | Description |
316
+ |------|-------------|
317
+ | `qwen-image-edit-2511-multiple-angles-lora.safetensors` | LoRA weights |
318
+ | `comfyui-workflow-multiple-angles.json` | ComfyUI workflow |
319
+
320
+ ▶️ [**Try it live on fal.ai**](https://fal.ai/models/fal-ai/qwen-image-edit-2511-multiple-angles)
321
+
322
+ ---
323
+
324
+ ## Recommended Settings
325
+
326
+ - **LoRA Strength**: 0.8 - 1.0
327
+ - **Base Model**: [Qwen/Qwen-Image-Edit-2511](https://huggingface.co/Qwen/Qwen-Image-Edit-2511)
328
+
329
+ ---
330
+
331
+ ## Code Examples
332
+
333
+ ### Python with Diffusers
334
+
335
+ ```python
336
+ import torch
337
+ from PIL import Image
338
+ from diffusers import QwenImageEditPlusPipeline
339
+
340
+ # Load base model
341
+ pipeline = QwenImageEditPlusPipeline.from_pretrained(
342
+ "Qwen/Qwen-Image-Edit-2511",
343
+ torch_dtype=torch.bfloat16
344
+ )
345
+ pipeline.to('cuda')
346
+
347
+ # Load Multi-Angles LoRA
348
+ pipeline.load_lora_weights("fal/Qwen-Image-Edit-2511-Multiple-Angles-LoRA")
349
+
350
+ # Generate
351
+ image = Image.open("input.png")
352
+ prompt = "<sks> front-right quarter view low-angle shot close-up"
353
+
354
+ output = pipeline(
355
+ image=image,
356
+ prompt=prompt,
357
+ generator=torch.manual_seed(0),
358
+ true_cfg_scale=4.0,
359
+ num_inference_steps=40,
360
+ guidance_scale=1.0,
361
+ )
362
+
363
+ output.images[0].save("output.png")
364
+ ```
365
+
366
+ ### Python with fal.ai
367
+
368
+ ```python
369
+ import fal_client
370
+
371
+ result = fal_client.subscribe(
372
+ "fal-ai/qwen-image-edit",
373
+ arguments={
374
+ "image_url": "https://your-image-url.jpg",
375
+ "prompt": "<sks> back view high-angle shot wide shot",
376
+ "lora_path": "fal/Qwen-Image-Edit-2511-Multiple-Angles-LoRA",
377
+ "lora_scale": 0.9,
378
+ },
379
+ )
380
+ print(result["images"][0]["url"])
381
+ ```
382
+
383
+ ---
384
+
385
+ ## Training Details
386
+
387
+ | Parameter | Value |
388
+ |-----------|-------|
389
+ | **Training Platform** | [fal.ai Qwen Image Edit 2511 Trainer](https://fal.ai/models/fal-ai/qwen-image-edit-2511-trainer) |
390
+ | **Base Model** | [Qwen/Qwen-Image-Edit-2511](https://huggingface.co/Qwen/Qwen-Image-Edit-2511) |
391
+ | **Training Data** | 3000+ Gaussian Splatting renders |
392
+ | **Camera Poses** | 96 unique positions (4×8×3) |
393
+ | **Data Source** | Synthetic 3D renders with precise camera control |
394
+ | **Dataset & Training** | Created by [@lovis93](https://huggingface.co/lovis93) ([@odinlovis](https://x.com/odinlovis)) |
395
+
396
+ ---
397
+
398
+ ## Tips for Best Results
399
+
400
+ 1. **Use the exact prompt format** - `<sks>` trigger is essential
401
+ 2. **Respect the order** - `[azimuth] [elevation] [distance]`
402
+ 3. **Start with LoRA strength 0.9** - Adjust based on results
403
+ 4. **Try low-angle shots** - This LoRA excels at low camera positions (-30°)
404
+ 5. **Input image matters** - Clear subjects with good lighting work best
405
+
406
+ ---
407
+
408
+ ## Related Work
409
+
410
+ - [Qwen-Image-Edit-2511](https://huggingface.co/Qwen/Qwen-Image-Edit-2511) - Base model
411
+ - [Flux-2-Multi-Angles-LoRA-v2](https://huggingface.co/lovis93/Flux-2-Multi-Angles-LoRA-v2) - Multi-angle for Flux (72 poses)
412
+ - [next-scene-qwen-image-lora](https://huggingface.co/lovis93/next-scene-qwen-image-lora)
413
+
414
+ ---
415
+
416
+ ## Author
417
+
418
+ Created by **Lovis Odin** — [@lovis93](https://huggingface.co/lovis93) • [@odinlovis](https://x.com/odinlovis)
419
+
420
+ Dataset created and model trained using [fal.ai Qwen Image Edit 2511 Trainer](https://fal.ai/models/fal-ai/qwen-image-edit-2511-trainer)
421
+
422
+ ---
423
+
424
+ **If you find this useful, please star the repo!**
425
+
426
+ **Issues?** Open a discussion in the Community tab