AI-Powered 2D Game Asset
Generation

When a user submits a prompt, the API route injects the prompt text, a random seed, and target dimensions into the workflow JSON template.

 1function buildWorkflow(prompt: string, seed: number, width: number, height: number) {
 2  const workflow = structuredClone(FLUX_WORKFLOW_TEMPLATE);
 3
 4  // Inject user prompt into CLIP text encode node
 5  workflow["4"].inputs.text = prompt;
 6
 7  // Set random seed for reproducibility
 8  workflow["6"].inputs.seed = seed;
 9
10  // Set output dimensions
11  workflow["3"].inputs.width = width;
12  workflow["3"].inputs.height = height;
13
14  return workflow;
15}

The built workflow is sent to ComfyUI's /prompt endpoint. ComfyUI returns a prompt_id that we use to track the generation job.

1const response = await fetch(`${{COMFYUI_URL}}/prompt`, {
2  method: 'POST',
3  headers: { 'Content-Type': 'application/json' },
4  body: JSON.stringify({ prompt: workflow })
5});
6
7const { prompt_id } = await response.json();

ComfyUI processes jobs asynchronously. We poll the /history endpoint until the job completes.

 1async function waitForCompletion(promptId: string): Promise<any> {
 2  while (true) {
 3    const res = await fetch(`${{COMFYUI_URL}}/history/${{promptId}}`);
 4    const data = await res.json();
 5
 6    if (data[promptId]) {
 7      return data[promptId];
 8    }
 9
10    // Wait before polling again
11    await new Promise(resolve => setTimeout(resolve, 1000));
12  }
13}

Once complete, we extract the output filename from the history response and fetch the image from ComfyUI's /view endpoint.

1const outputs = history.outputs;
2const imageNode = outputs["9"]; // Save Image node
3const { filename, subfolder, type } = imageNode.images[0];
4
5const imageUrl = `${{COMFYUI_URL}}/view?filename=${{filename}}&subfolder=${{subfolder}}&type=${{type}}`;
6const imageResponse = await fetch(imageUrl);