@dhee_ai/bundle-infographics

Narrated infographic videos: Ideogram 4 infographic stills, Qwen3-TTS cloned-voice narration, crisp ffmpeg Ken Burns motion + crossfades.

A Dhee bundle (pipeline) providing the bundle id infographics. Discovered by dhee-core via the dhee-bundle-* npm convention; referenced as npm:@dhee_ai/bundle-infographics#infographics.

Pipeline

brief → outline → infographic_plan ─┬─ segment_image_prompt → segment_image   (Ideogram 4 / comfy.tti)
                                    │
                  narrator_voice ───┴─ segment_narration → segment_audio      (Qwen3-TTS / comfy.tts, cloned voice)
                                                                  │
                  segment_image + segment_audio → segment_video   (ffmpeg.kenburns — pixel-sharp pan/zoom, sized to narration)
                                                                  │
                                                          final_video          (ffmpeg.concat, xfade crossfades)

The LLM distills the brief into 4–7 segments. Each becomes one rich Ideogram 4 infographic still — pure data design, no people or talking heads — and one Qwen3-TTS narration track in one consistent voice (a narrator voice is designed + saved once by narrator_voice, then cloned per segment). Each still is animated with a subtle, pixel-sharp ffmpeg Ken Burns move sized to its narration (generative video mangles crisp text/charts), and the clips are concatenated with crossfade transitions. Audio is generated first so each clip is sized to its narration.

Requirements

• Runners: llm.generate, comfy.tti, ffmpeg.kenburns, ffmpeg.concat (built into dhee-core) and comfy.tts — install dhee-runner-tts (declared under dependencies.runnerPackages).
• ComfyUI with the Ideogram 4 and Qwen3-TTS (FB_Qwen3TTS*) custom nodes, reachable at your self.local endpoint (COMFYUI_BASE_URL / ENDPOINT_self_local).

Drop-in workflows (per machine)

The shipped workflows/ files target a reference box — replace with your machine's exported API-format graphs if node ids differ:

• workflows/ideogram4.json (+ ideogram4.manifest.json) — keep the manifest parameterMappings node ids aligned with the graph (prompt → 98:24, seed, filename).
• workflows/qwen3_voice_save.json — designs + saves the Narrator speaker once (narrator_voice stage).
• workflows/qwen3_narration_cloned.json — clones Narrator per segment (FB_Qwen3TTSLoadSpeaker → RoleBank → DialogueInference); the comfy.tts runner injects Narrator: <text> into the script node.

Layout

bundles/infographics/
  bundle.json     # the DAG (id, goal, nodes, dependencies)
  inputs/brief.md # default brief (replace per project)
  prompts/        # LLM prompt templates
  schemas/        # JSON schemas for json-output nodes
  workflows/      # ComfyUI workflows + comfy.tti manifest

License

MIT