Skip to content
GenMedia Creative Studio

Creative Workflows

Beyond single-model workspaces, GenMedia Creative Studio features Creative Workflows—opinionated, multi-stage pipelines that orchestrate Gemini reasoning, Nano Banana image synthesis, Virtual Try-On (VTO), Veo video generation, and MoviePy post-processing to solve end-to-end industry use cases.

Graduated Creative Workflows

Section titled “Graduated Creative Workflows”
Workflow Route Models & Tools Orchestrated End-to-End Pipeline Summary
Storyboarder /storyboarder Gemini 2.5/3.5 Flash (NarrativeStructure JSON) + Nano Banana + Veo + MoviePy Expands a story concept and optional character reference into a structured multi-scene script, generates consistent scene keyframes, animates each scene with Veo, and stitches a final master MP4 movie.
Shop the Look /shop-the-look Gemini + Vertex AI VTO (virtual-try-on-001) + Gemini Critic + Veo Enriches apparel catalog articles, sequentially dresses a virtual model item-by-item, uses a Gemini visual critic to select the best sample and retry failed garments, and generates a Veo runway video.
Character Consistency /character-consistency Gemini (FacialCompositeProfile & BestImage) + Imagen 3 / Nano Banana + Veo Extracts a structured facial profile from multi-angle character photos, synthesizes candidate scenes, uses Gemini to select the most likeness-faithful candidate, outpaints to 16:9, and animates with Veo.
Interior Design /interior-design Gemini + Nano Banana (gemini-2.5-flash-image) + Veo 3.1 + MoviePy Converts a 2D floor plan into a 3D isometric visualization, extracts room names, generates zoomed first-person views per room, supports conversational material re-styling, and stitches a guided Veo video walkthrough.
Virtual Try-On (VTO) /vto VirtualModelGenerator + Vertex AI VTO (recontext_image) + Veo Combines a real or synthetically generated model portrait with up to 4 product garments for photorealistic try-on synthesis and one-click Veo motion animation.
Starter Pack /starter-pack Nano Banana (1:1) + VirtualModelGenerator Bi-directional retail moodboard pipeline: extracts a “Starter Pack” flat-lay product collage from a lifestyle look, or dresses a virtual model in all items from a Starter Pack collage.
Motion Portraits /motion-portraits Live selfie_camera / Library + Gemini (create_scene_direction) + Veo + MoviePy Captures or imports a portrait, applies curated cinematic style modifiers (PORTRAIT_STYLES), optionally expands the motion direction via Gemini, and generates both a Veo MP4 clip and an animated GIF.
Pixie Compositor /pixie-compositor MoviePy (concatenate_videoclips, CompositeVideoClip, write_gif) Post-production assembly tool for concatenating multiple Veo clips with transitions (concat, crossfade, wipe), layering Lyria or TTS audio tracks onto video, and exporting looping GIFs.

Experimental & Labs Workflows

Section titled “Experimental & Labs Workflows”

In addition to the graduated workflows in the main navigation bar, GenMedia Creative Studio includes an experimental Labs & Experimental Workflows hub (/labs) where domain-specific prototypes—such as Retro Games (/retro_games), Brand Adherence (/brand_adherence), Character Asset Sheet (/character_sheet), Guideline Analysis (/guideline-analysis), and Object Rotation (/object-rotation)—are incubated before graduating to the main navigation.

Common Orchestration Patterns

Section titled “Common Orchestration Patterns”

Across all Creative Workflows, the studio uses four recurring architectural patterns:

  1. Structured Gemini Planning & Prompt Expansion (response_schema): Workflows use Gemini with Pydantic schemas (NarrativeStructure, FacialCompositeProfile, RoomNames, TryOnOrder) to decompose a high-level user brief into deterministic downstream model calls.
  2. Multi-Turn Visual Context Carryover: Workflows like Storyboarder, Shop the Look, and Interior Design feed the generated image from Step $N$ directly into Step $N+1$ as a reference image so characters, garments, and architectural layouts remain coherent.
  3. Agentic Visual Judge & Critic Loops: Workflows like Shop the Look and Character Consistency generate $N$ candidate images in parallel and invoke a Gemini multimodal judge (select_best_image(), final_image_critic()) to score garment accuracy and facial likeness, automatically retrying failed steps.
  4. Streaming Generator Progress & Firestore Checkpoint Persistence: Every workflow executes inside a Python generator (yield) that updates the Mesop UI after each sub-step and logs final MediaItem records to Firestore (genmedia_library).