ExternalVendorCode/LocalAI
1
0
Fork 0
mirror of https://github.com/mudler/LocalAI.git synced 2024-12-20 05:07:54 +00:00
Code Issues Actions 13 Packages Projects Releases Wiki Activity

fix: guidance_scale not work in sd (#1488)

Browse Source 
Signed-off-by: hibobmaster <32976627+hibobmaster@users.noreply.github.com>
This commit is contained in:
BobMaster 2023-12-25 02:24:52 +08:00 committed by GitHub
parent 6597881854
commit 7e2d101a46
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 19 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

41
backend/python/diffusers/backend_diffusers.py
View File

@ -149,9 +149,9 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
local = False local = False
modelFile = request.Model modelFile = request.Model
cfg_scale = 7 self.cfg_scale = 7
if request.CFGScale != 0: if request.CFGScale != 0:
cfg_scale = request.CFGScale self.cfg_scale = request.CFGScale
clipmodel = "runwayml/stable-diffusion-v1-5" clipmodel = "runwayml/stable-diffusion-v1-5"
if request.CLIPModel != "": if request.CLIPModel != "":
@ -173,17 +173,14 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
if (request.PipelineType == "StableDiffusionImg2ImgPipeline") or (request.IMG2IMG and request.PipelineType == ""): if (request.PipelineType == "StableDiffusionImg2ImgPipeline") or (request.IMG2IMG and request.PipelineType == ""):
if fromSingleFile: if fromSingleFile:
self.pipe = StableDiffusionImg2ImgPipeline.from_single_file(modelFile, self.pipe = StableDiffusionImg2ImgPipeline.from_single_file(modelFile,
torch_dtype=torchType, torch_dtype=torchType)
guidance_scale=cfg_scale)
else: else:
self.pipe = StableDiffusionImg2ImgPipeline.from_pretrained(request.Model, self.pipe = StableDiffusionImg2ImgPipeline.from_pretrained(request.Model,
torch_dtype=torchType, torch_dtype=torchType)
guidance_scale=cfg_scale)
elif request.PipelineType == "StableDiffusionDepth2ImgPipeline": elif request.PipelineType == "StableDiffusionDepth2ImgPipeline":
self.pipe = StableDiffusionDepth2ImgPipeline.from_pretrained(request.Model, self.pipe = StableDiffusionDepth2ImgPipeline.from_pretrained(request.Model,
torch_dtype=torchType, torch_dtype=torchType)
guidance_scale=cfg_scale)
## img2vid ## img2vid
elif request.PipelineType == "StableVideoDiffusionPipeline": elif request.PipelineType == "StableVideoDiffusionPipeline":
self.img2vid=True self.img2vid=True
@ -197,38 +194,32 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
self.pipe = AutoPipelineForText2Image.from_pretrained(request.Model, self.pipe = AutoPipelineForText2Image.from_pretrained(request.Model,
torch_dtype=torchType, torch_dtype=torchType,
use_safetensors=SAFETENSORS, use_safetensors=SAFETENSORS,
variant=variant, variant=variant)
guidance_scale=cfg_scale)
elif request.PipelineType == "StableDiffusionPipeline": elif request.PipelineType == "StableDiffusionPipeline":
if fromSingleFile: if fromSingleFile:
self.pipe = StableDiffusionPipeline.from_single_file(modelFile, self.pipe = StableDiffusionPipeline.from_single_file(modelFile,
torch_dtype=torchType, torch_dtype=torchType)
guidance_scale=cfg_scale)
else: else:
self.pipe = StableDiffusionPipeline.from_pretrained(request.Model, self.pipe = StableDiffusionPipeline.from_pretrained(request.Model,
torch_dtype=torchType, torch_dtype=torchType)
guidance_scale=cfg_scale)
elif request.PipelineType == "DiffusionPipeline": elif request.PipelineType == "DiffusionPipeline":
self.pipe = DiffusionPipeline.from_pretrained(request.Model, self.pipe = DiffusionPipeline.from_pretrained(request.Model,
torch_dtype=torchType, torch_dtype=torchType)
guidance_scale=cfg_scale)
elif request.PipelineType == "VideoDiffusionPipeline": elif request.PipelineType == "VideoDiffusionPipeline":
self.txt2vid=True self.txt2vid=True
self.pipe = DiffusionPipeline.from_pretrained(request.Model, self.pipe = DiffusionPipeline.from_pretrained(request.Model,
torch_dtype=torchType, torch_dtype=torchType)
guidance_scale=cfg_scale)
elif request.PipelineType == "StableDiffusionXLPipeline": elif request.PipelineType == "StableDiffusionXLPipeline":
if fromSingleFile: if fromSingleFile:
self.pipe = StableDiffusionXLPipeline.from_single_file(modelFile, self.pipe = StableDiffusionXLPipeline.from_single_file(modelFile,
torch_dtype=torchType, use_safetensors=True, torch_dtype=torchType,
guidance_scale=cfg_scale) use_safetensors=True)
else: else:
self.pipe = StableDiffusionXLPipeline.from_pretrained( self.pipe = StableDiffusionXLPipeline.from_pretrained(
request.Model, request.Model,
torch_dtype=torchType, torch_dtype=torchType,
use_safetensors=True, use_safetensors=True,
variant=variant, variant=variant)
guidance_scale=cfg_scale)
if CLIPSKIP and request.CLIPSkip != 0: if CLIPSKIP and request.CLIPSkip != 0:
self.clip_skip = request.CLIPSkip self.clip_skip = request.CLIPSkip
@ -384,12 +375,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
image = image.resize((1024, 576)) image = image.resize((1024, 576))
generator = torch.manual_seed(request.seed) generator = torch.manual_seed(request.seed)
frames = self.pipe(image, decode_chunk_size=CHUNK_SIZE, generator=generator).frames[0] frames = self.pipe(image, guidance_scale=self.cfg_scale, decode_chunk_size=CHUNK_SIZE, generator=generator).frames[0]
export_to_video(frames, request.dst, fps=FPS) export_to_video(frames, request.dst, fps=FPS)
return backend_pb2.Result(message="Media generated successfully", success=True) return backend_pb2.Result(message="Media generated successfully", success=True)
if self.txt2vid: if self.txt2vid:
video_frames = self.pipe(prompt, num_inference_steps=steps, num_frames=int(FRAMES)).frames video_frames = self.pipe(prompt, guidance_scale=self.cfg_scale, num_inference_steps=steps, num_frames=int(FRAMES)).frames
export_to_video(video_frames, request.dst) export_to_video(video_frames, request.dst)
return backend_pb2.Result(message="Media generated successfully", success=True) return backend_pb2.Result(message="Media generated successfully", success=True)
@ -399,12 +390,14 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
kwargs["prompt_embeds"]= conditioning kwargs["prompt_embeds"]= conditioning
# pass the kwargs dictionary to the self.pipe method # pass the kwargs dictionary to the self.pipe method
image = self.pipe( image = self.pipe(
guidance_scale=self.cfg_scale,
**kwargs **kwargs
).images[0] ).images[0]
else: else:
# pass the kwargs dictionary to the self.pipe method # pass the kwargs dictionary to the self.pipe method
image = self.pipe( image = self.pipe(
prompt, prompt,
guidance_scale=self.cfg_scale,
**kwargs **kwargs
).images[0] ).images[0]