project_6/upstream_ref/xllm/xllm/proto/image_generation.proto

syntax = "proto3";

option go_package = "jd.com/jd-infer/xllm;xllm";
package xllm.proto;

import "common.proto";
import "tensor.proto";

// Input parameters container
message Input {
  // Primary input text description for image generation
  string prompt = 1;

  // Secondary prompt for additional details (e.g., color, lighting)
  optional string prompt_2 = 2;

  // Negative prompt to exclude low-quality features
  optional string negative_prompt = 3;

  // Secondary negative prompt to exclude additional unwanted features
  optional string negative_prompt_2 = 4;

  // prompt embedding
  optional Tensor prompt_embed = 5;

  // pooled prompt embeddings
  optional Tensor pooled_prompt_embed = 6;

  // negative prompt embedding: [num_prompts][embedding_dim]
  optional Tensor negative_prompt_embed = 7;

  //pooled negative prompt embedding
  optional Tensor negative_pooled_prompt_embed = 8;

  // initial latent: [batch_size][channels][height/8][width/8]
  optional Tensor latent = 9;

  // Input type: "base64"
  optional string image = 10;

  // Input type: "base64"
  optional string mask_image = 11;

  // An image batch of mask images generated by the VAE
  optional Tensor masked_image_latent = 12;

  // Control Image
  optional string control_image = 13;

  // Condition Image
  optional string condition_image = 14;
}

// Generation parameters container
message Parameters {
  // Size of the generated image in pixels, default in 1024*1024
  optional string size = 1;

  // Number of inference steps for image generation
  optional int32 num_inference_steps = 2;

  // True CFG scale value for balancing generation
  optional float true_cfg_scale = 3;

  // Guidance scale value for prompt adherence
  optional float guidance_scale = 4;

  // Number of images to generate per prompt
  optional int32 num_images_per_prompt = 5;

  // Random seed value for image generation
  optional int64 seed = 6;

  // Maximum sequence length for prompt processing
  optional int32 max_sequence_length = 7;

  // The extent to which the reference image is altered, between 0 and 1
  optional float strength = 8;

  // Array of sigma values for noise scheduling
  // repeated float sigmas = 9;

  // Enable CFG renormalization for improved stability
  optional bool enable_cfg_renorm = 10;

  // Minimum value for CFG renorm scale
  optional float cfg_renorm_min = 11;

  // Output type, either "base64" or "url"
  // optional string output_type = 12;
}

// Request structure for image generation tasks using FLUX models
message ImageGenerationRequest {
  // ID of the FLUX text-to-image model to use. Currently supported values are "flux-schnell" and "flux-dev".
  string model = 1;

  Input input = 2;

  Parameters parameters = 3;

  // Unique identifier representing the end-user
  optional string user = 4;

  // ID used by server to identify the request for tracking and troubleshooting
  optional string request_id = 5;
}

// Individual image generation result data
message ImageGenData {

  optional string image = 1;

  // Width of the generated image in pixels
  int32 width = 3;

  // Height of the generated image in pixels
  int32 height = 4;

  // Seed used for generating this image
  int64 seed = 5;
}

// Output container for image generation task results
message ImageGenerationOutput {

  // List of generated image data
  repeated ImageGenData results = 1;
}

// Response structure for image generation requests
message ImageGenerationResponse {
  // The ID of the response.
  string id = 1;

  // The object type of the image response.
  // [default = "list"]
  string object = 2;

  // The Unix timestamp of when the embedding response was created.
  int64 created = 3;

  // The model used to generate the embedding response.
  string model = 4;

  // Contains task details and generation results
  ImageGenerationOutput output = 5;
}