Ti depicting Alice from MC. A pretty model with fine face and figure. Heavy NSFW element in the training data, so a negative NSFW/nude prompt may be necessary for SFW generations.
All sample images use ADetailer (face) and Highres. fix (x2); no other embeddings, LoRa or manual inpaints were used.
TI was trained on a dataset of 80 publicly-available images with a batch size of 8 and gradient accumulation of 10 (80 steps per epoch) for 330 epochs using DADAPT-LION with constant LR of 0.8. Image masking was leveraged to focus the training.
All models depicted in training data are over the age of eighteen years at the time of the creation of such depictions per 18 U.S.C. § 2257.
Description
Version 2.0.
TI was trained on a dataset of 80 publicly-available images with a batch size of 8 and gradient accumulation of 10 (80 steps per epoch) for 330 epochs using DADAPT-LION with constant LR of 0.8. Image masking was leveraged to focus the training.
All models depicted in training data are over the age of eighteen years at the time of the creation of such depictions per 18 U.S.C. § 2257.
FAQ
Comments (10)
v2 is looking much better! Nice work!
Trying to use this on Automatic1111 generates this error:
*** Error completing request *** Arguments: ('task(1btdjyw96m5bini)', "(Color photograph) of (very beautiful) woman, alice03, (girl's bedroom). candid, (full color), face detail, intricate high detail, dramatic, skin pores, cinematic lighting, detailed, (vibrant, photo realistic, dramatic, sharp focus) ((film grain, skin details, high detailed skin texture, 8k hdr, dslr))", '', ['Schizo Negative'], 30, 'Euler a', 1, 1, 7, 768, 512, False, 0.7, 2, 'Latent', 0, 0, 0, 'Use same checkpoint', 'Use same sampler', '', '', [], <gradio.routes.Request object at 0x7ef3b5f41240>, 0, False, '', 0.8, -1, False, -1, 0, 0, 0, False, False, 'positive', 'comma', 0, False, False, '', 1, '', [], 0, '', [], 0, '', [], True, False, False, False, 0, False) {} Traceback (most recent call last): File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/call_queue.py", line 57, in f res = list(func(*args, **kwargs)) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/call_queue.py", line 36, in f res = func(*args, kwargs) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/txt2img.py", line 55, in txt2img processed = processing.process_images(p) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/processing.py", line 732, in process_images res = process_images_inner(p) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/processing.py", line 867, in process_images_inner samples_ddim = p.sample(conditioning=p.c, unconditional_conditioning=p.uc, seeds=p.seeds, subseeds=p.subseeds, subseed_strength=p.subseed_strength, prompts=p.prompts) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/processing.py", line 1140, in sample samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x)) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/sd_samplers_kdiffusion.py", line 235, in sample samples = self.launch_sampling(steps, lambda: self.func(self.model_wrap_cfg, x, extra_args=self.sampler_extra_args, disable=False, callback=self.callback_state, extra_params_kwargs)) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/sd_samplers_common.py", line 261, in launch_sampling return func() File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/sd_samplers_kdiffusion.py", line 235, in <lambda> samples = self.launch_sampling(steps, lambda: self.func(self.model_wrap_cfg, x, extra_args=self.sampler_extra_args, disable=False, callback=self.callback_state, **extra_params_kwargs)) File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 115, in decorate_context return func(*args, **kwargs) File "/content/gdrive/MyDrive/sd/stablediffusion/src/k-diffusion/k_diffusion/sampling.py", line 145, in sample_euler_ancestral denoised = model(x, sigmas[i] s_in, *extra_args) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1518, in wrappedcall_impl return self._call_impl(*args, **kwargs) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1527, in callimpl return forward_call(*args, kwargs) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/sd_samplers_cfg_denoiser.py", line 201, in forward devices.test_for_nans(x_out, "unet") File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/devices.py", line 136, in test_for_nans raise NansException(message) modules.devices.NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check. --- 100% 30/30 [00:09<00:00, 3.10it/s] 3% 1/30 [00:01<00:48, 1.68s/it] * Error completing request *** Arguments: ('task(op0acz5nkihmf4k)', "(Color photograph) of (very beautiful) woman, alice03, (perfect natural breasts), (erect nipples), (girl's bedroom). candid, (full color), face detail, intricate high detail, dramatic, skin pores, cinematic lighting, detailed, (vibrant, photo realistic, dramatic, sharp focus) ((film grain, skin details, high detailed skin texture, 8k hdr, dslr))", '', ['Schizo Negative'], 30, 'Euler a', 1, 1, 7, 768, 512, False, 0.7, 2, 'Latent', 0, 0, 0, 'Use same checkpoint', 'Use same sampler', '', '', [], <gradio.routes.Request object at 0x7ef3b4dbd180>, 0, False, '', 0.8, -1, False, -1, 0, 0, 0, False, False, 'positive', 'comma', 0, False, False, '', 1, '', [], 0, '', [], 0, '', [], True, False, False, False, 0, False) {} Traceback (most recent call last): File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/call_queue.py", line 57, in f res = list(func(*args, **kwargs)) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/call_queue.py", line 36, in f res = func(*args, kwargs) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/txt2img.py", line 55, in txt2img processed = processing.process_images(p) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/processing.py", line 732, in process_images res = process_images_inner(p) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/processing.py", line 867, in process_images_inner samples_ddim = p.sample(conditioning=p.c, unconditional_conditioning=p.uc, seeds=p.seeds, subseeds=p.subseeds, subseed_strength=p.subseed_strength, prompts=p.prompts) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/processing.py", line 1140, in sample samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x)) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/sd_samplers_kdiffusion.py", line 235, in sample samples = self.launch_sampling(steps, lambda: self.func(self.model_wrap_cfg, x, extra_args=self.sampler_extra_args, disable=False, callback=self.callback_state, extra_params_kwargs)) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/sd_samplers_common.py", line 261, in launch_sampling return func() File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/sd_samplers_kdiffusion.py", line 235, in <lambda> samples = self.launch_sampling(steps, lambda: self.func(self.model_wrap_cfg, x, extra_args=self.sampler_extra_args, disable=False, callback=self.callback_state, **extra_params_kwargs)) File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 115, in decorate_context return func(*args, **kwargs) File "/content/gdrive/MyDrive/sd/stablediffusion/src/k-diffusion/k_diffusion/sampling.py", line 145, in sample_euler_ancestral denoised = model(x, sigmas[i] s_in, *extra_args) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1518, in wrappedcall_impl return self._call_impl(*args, **kwargs) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1527, in callimpl return forward_call(*args, kwargs) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/sd_samplers_cfg_denoiser.py", line 201, in forward devices.test_for_nans(x_out, "unet") File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/devices.py", line 136, in test_for_nans raise NansException(message) modules.devices.NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check. --- 3% 1/30 [00:00<00:18, 1.57it/s] * Error completing request *** Arguments: ('task(u1oess9pbadavx1)', "(Color photograph) of (very beautiful) woman, alice03, (perfect natural breasts), (erect nipples), (girl's bedroom). candid, (full color), face detail, intricate high detail, dramatic, skin pores, cinematic lighting, detailed, (vibrant, photo realistic, dramatic, sharp focus) ((film grain, skin details, high detailed skin texture, 8k hdr, dslr))", '', ['Schizo Negative'], 30, 'Euler a', 1, 1, 7, 768, 512, False, 0.7, 2, 'Latent', 0, 0, 0, 'Use same checkpoint', 'Use same sampler', '', '', [], <gradio.routes.Request object at 0x7ef39ef4f370>, 0, False, '', 0.8, -1, False, -1, 0, 0, 0, False, False, 'positive', 'comma', 0, False, False, '', 1, '', [], 0, '', [], 0, '', [], True, False, False, False, 0, False) {} Traceback (most recent call last): File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/call_queue.py", line 57, in f res = list(func(*args, **kwargs)) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/call_queue.py", line 36, in f res = func(*args, kwargs) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/txt2img.py", line 55, in txt2img processed = processing.process_images(p) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/processing.py", line 732, in process_images res = process_images_inner(p) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/processing.py", line 867, in process_images_inner samples_ddim = p.sample(conditioning=p.c, unconditional_conditioning=p.uc, seeds=p.seeds, subseeds=p.subseeds, subseed_strength=p.subseed_strength, prompts=p.prompts) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/processing.py", line 1140, in sample samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x)) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/sd_samplers_kdiffusion.py", line 235, in sample samples = self.launch_sampling(steps, lambda: self.func(self.model_wrap_cfg, x, extra_args=self.sampler_extra_args, disable=False, callback=self.callback_state, extra_params_kwargs)) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/sd_samplers_common.py", line 261, in launch_sampling return func() File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/sd_samplers_kdiffusion.py", line 235, in <lambda> samples = self.launch_sampling(steps, lambda: self.func(self.model_wrap_cfg, x, extra_args=self.sampler_extra_args, disable=False, callback=self.callback_state, **extra_params_kwargs)) File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 115, in decorate_context return func(*args, **kwargs) File "/content/gdrive/MyDrive/sd/stablediffusion/src/k-diffusion/k_diffusion/sampling.py", line 145, in sample_euler_ancestral denoised = model(x, sigmas[i] s_in, *extra_args) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1518, in wrappedcall_impl return self._call_impl(*args, **kwargs) File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1527, in callimpl return forward_call(*args, **kwargs) File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/sd_samplers_cfg_denoiser.py", line 201, in forward devices.test_for_nans(x_out, "unet") File "/content/gdrive/MyDrive/sd/stable-diffusion-webui/modules/devices.py", line 136, in test_for_nans raise NansException(message) modules.devices.NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
Tried running it with the float32 option and still gave this error, and I haven't had this error with other TIs in the form of safetensors either.
I believe the answer is in the error message, but may want to reach out to the Automatic1111 team to confirm; if you haven't restarted your A1111 instance from the command line after making the config changes, I'd recommend doing so. I'd also confirm you've placed the embedding in the appropriate folder (stable-diffusion-webui\embeddings) and not in a LoRa folder somewhere.:
A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check
As a final note, the SHA-256 hash of the embedding on my machine (subsequently uploaded to CivitAi) is 214075B8EF2D33AD341FA4B915B2D3C0370E390AC50FDC296E8043DBB32C8244 (powershell command: get-filehash filename.safetensors). If your copy differs, I'd recommend redownloading to ensure you don't have a broken file.
Good luck!
@sendmefreestuffa8246 I've tried running this embedding on multiple fresh restarts, so I don't think that's it. It's definitely in the embedding folder (I don't really use LoRas), and I'm not really using the command line version of the software.
@HaikenEdge Amusingly, I was able to reproduce the error when I converted this embedding from fp32 (the precision uploaded to Civitai) to fp16. I'm not certain if you have any automatic processes/extensions (batch application of the Automatic1111 extension ModelConverter, perhaps?) that converts models to fp16, but I'd recommend redownloading the fp32 embedding from Civitai and retrying.
I'm hesitant to upload the fp16 version I have because I can't test it in a known-good environment; I'd prefer not to have probably-broken models uploaded.
@sendmefreestuffa8246 I have no extensions installed beyond what's built into Automatic1111, and redownloading the embedding from Civitai hasn't fixed the problem.
@HaikenEdge I converted the safetensors versions to pickle and uploaded that. Check the versions up top.
While I'm hoping it resolves your issues, I rather suspect your environment isn't handling the fp32 precision of the embedding well. If that's the case, the pickle will result in the same error.
Fingers crossed!
@sendmefreestuffa8246 The pickle doesn't cause the error, but it also doesn't generate an image matching the examples given here; see the included image in the gallery.
@HaikenEdge At least it's generating! For the sample images in my gallery, I'm using highres. fix x2 and ADetailer (face); I think most are a starting resolution of 384x512 upscaled to 768x1024. I've also found that photoreal models provide better facial fidelity (at the cost of stylistic flourish, naturally).
@HaikenEdge Ugh. Just found out it wasn't loaded on my system either (I don't have unsafe pickle loading enabled), so the generated sample images are actually just using the original fp32 embedding with the text token "_fp16" modifying it slightly.
I'm not going to try converting the fp32 version any more. I'm midway through a fp16 train using the original dataset. I've changed up the mix of the images a bit, with 27x2 (repeats) face and 42 body for a total of 96 images per step. I'll post that once it's done, I've selected the best output and generated some sample images.
I'm also removing the "pickle" version of this model; it may work, but I can't test it (same issue I had with the last convert), so I don't want it posted.
Standby.