fix(diffusers): correctly parse height and width request without parametrization (#4082)

* fix(diffusers): allow to specify width and height without enable-parameters

Let's simplify usage by not gating width and height by parameters

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

* chore: use sane defaults

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

---------

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>
This commit is contained in:
Ettore Di Giacinto 2024-11-06 08:53:02 +01:00 committed by GitHub
parent b59e16742e
commit b425a870b0
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 14 additions and 4 deletions

View File

@ -409,8 +409,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
# create a dictionary of values for the parameters
options = {
"negative_prompt": request.negative_prompt,
"width": request.width,
"height": request.height,
"num_inference_steps": steps,
}
@ -428,13 +426,13 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
keys = options.keys()
if request.EnableParameters != "":
keys = request.EnableParameters.split(",")
keys = [key.strip() for key in request.EnableParameters.split(",")]
if request.EnableParameters == "none":
keys = []
# create a dictionary of parameters by using the keys from EnableParameters and the values from defaults
kwargs = {key: options[key] for key in keys}
kwargs = {key: options.get(key) for key in keys if key in options}
# Set seed
if request.seed > 0:
@ -445,6 +443,12 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
if self.PipelineType == "FluxPipeline":
kwargs["max_sequence_length"] = 256
if request.width:
kwargs["width"] = request.width
if request.height:
kwargs["height"] = request.height
if self.PipelineType == "FluxTransformer2DModel":
kwargs["output_type"] = "pil"
kwargs["generator"] = torch.Generator("cpu").manual_seed(0)
@ -464,6 +468,7 @@ class BackendServicer(backend_pb2_grpc.BackendServicer):
export_to_video(video_frames, request.dst)
return backend_pb2.Result(message="Media generated successfully", success=True)
print(f"Generating image with {kwargs=}", file=sys.stderr)
image = {}
if COMPEL:
conditioning, pooled = self.compel.build_conditioning_tensor(prompt)

View File

@ -136,6 +136,11 @@ func ImageEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, appCon
config.Backend = model.StableDiffusionBackend
}
if !strings.Contains(input.Size, "x") {
input.Size = "512x512"
log.Warn().Msgf("Invalid size, using default 512x512")
}
sizeParts := strings.Split(input.Size, "x")
if len(sizeParts) != 2 {
return fmt.Errorf("invalid value for 'size'")