making some changes to support basic text prompts

src/index.mts

@@ -3,7 +3,7 @@ import path from "node:path"
 
 import express from "express"
 
-import { VideoTask, VideoSequenceRequest } from "./types.mts"
+import { VideoTask, VideoTaskRequest } from "./types.mts"
 import { parseVideoRequest } from "./utils/parseVideoRequest.mts"
 import { savePendingTask } from "./scheduler/savePendingTask.mts"
 import { getTask } from "./scheduler/getTask.mts"
@@ -21,8 +21,8 @@ const port = 7860
 app.use(express.json())
 
 app.post("/", async (req, res) => {
-  const request = req.body as VideoSequenceRequest
-  
+  const request = req.body as VideoTaskRequest
+
   if (!hasValidAuthorization(req.headers)) {
     console.log("Invalid authorization")
     res.status(401)
@@ -30,8 +30,7 @@ app.post("/", async (req, res) => {
     res.end()
     return
   }
-
-
+  
   let task: VideoTask = null
 
   console.log(`creating task from request..`)
@@ -150,7 +149,7 @@ app.get("/video/:id\.mp4", async (req, res) => {
     return
   }
 
-  
+
   if (!req.params.id) {
     res.status(400)
     res.write(JSON.stringify({ error: "please provide a valid video id" }))

src/types.mts

@@ -219,10 +219,10 @@ export interface VideoSequenceData {
 
 export type VideoSequence = VideoSequenceMeta & VideoSequenceData
 
-export type VideoSequenceRequest = {
-  token: string
-  sequence: VideoSequenceMeta
-  shots: VideoShotMeta[]
+export type VideoTaskRequest = {
+  prompt: string
+  sequence: Partial<VideoSequenceMeta>
+  shots: Array<Partial<VideoShotMeta>>
 }
 
 export type VideoTask = VideoSequence & {

src/utils/parseVideoRequest.mts

@@ -1,20 +1,33 @@
 import { v4 as uuidv4 } from "uuid"
+import { HfInference } from "@huggingface/inference"
 
 // convert a request (which might be invalid)
 
-import { VideoSequenceRequest, VideoTask } from "../types.mts"
+import { VideoTaskRequest, VideoTask, VideoShotMeta } from "../types.mts"
 import { getValidNumber } from "./getValidNumber.mts"
 import { getValidResolution } from "./getValidResolution.mts"
 import { parseShotRequest } from "./parseShotRequest.mts"
 import { generateSeed } from "./generateSeed.mts"
 import { sequenceFormatVersion } from "../config.mts"
 
+// const hfi = new HfInference(process.env._VC_HF_API_TOKEN)
+// const hf = hfi.endpoint(process.env.VC_INFERENCE_ENDPOINT_URL)
 
-export const parseVideoRequest = async (request: VideoSequenceRequest): Promise<VideoTask> => {
+export const parseVideoRequest = async (request: VideoTaskRequest): Promise<VideoTask> => {
   // we don't want people to input their own ID or we might have trouble,
   // such as people attempting to use a non-UUID, a file path (to hack us), etc
   const id = uuidv4()
 
+  if (typeof request.prompt === "string" && request.prompt.length > 0) {
+      // TODO: use llama2 to populate this!
+    request.sequence = {
+      videoPrompt: request.prompt,
+    }
+    request.shots = [{
+      shotPrompt: request.prompt,
+    }]
+  }
+
   const task: VideoTask = {
     // ------------ VideoSequenceMeta -------------
     id,