Consistent AI Avatar LoRA + Video Workflow Setup

I’m looking for an experienced AI video creator / Stable Diffusion specialist to help build a consistent, cinematic AI avatar named Astra Loné for short fitness videos. (See attached full brief)

The goal is to:

Train a LoRA model of Astra Loné for character consistency.

Integrate that LoRA into a RunDiffusion (ComfyUI + AnimateDiff) workflow.

Produce short 10–15 second cinematic clips (e.g., plank, squat, push-up) that maintain the same look, outfit, lighting, and background across all videos.

Deliverables

LoRA Training

Use 6–10 supplied reference images of Astra Loné.

Train and test a LoRA (Astra_Loné_v1.safetensors) that replicates her appearance, outfit, and cinematic tone.

Provide the final LoRA file and brief training summary.

Workflow Setup

Build a ComfyUI + AnimateDiff workflow (either locally or on RunDiffusion).

Integrate the LoRA for consistent output.

Configure key nodes:

LoRA Loader

AnimateDiff Motion

(Optional) ControlNet OpenPose for body positioning.

Set render settings for warm cinematic lighting, soft shadows, and smooth movement.

Test Videos

Produce 2–3 short sample clips (10 seconds each):

Example actions: Plank, Squat, Stretch

Each clip should use the same Astra Loné avatar and visual style (luxury gym, black outfit, cinematic realism).

Handover

Deliver the complete ComfyUI workflow (.json) and render settings.

Provide written instructions for reusing and adapting the workflow for new exercises.

Ideal Skills & Experience

Proven experience with Stable Diffusion (SDXL or SD 1.5)

Skilled in ComfyUI, AnimateDiff, and ControlNet (OpenPose)

Hands-on knowledge of LoRA or DreamBooth training

Experience producing realistic human-style AI video

Familiarity with RunDiffusion, OpenArt, or cloud GPU setups

Attention to detail for character consistency and cinematic quality

Reference Keywords

LoRA training, Stable Diffusion, ComfyUI, AnimateDiff, RunDiffusion, ControlNet, OpenPose, AI video generation, cinematic lighting, AI avatar, short-form content, SDXL.

Screening Questions (Please Answer When Applying)

Have you personally trained and deployed a LoRA model before? If yes, please provide an example or link to your work.

Have you used ComfyUI or AnimateDiff to produce motion clips? Describe your workflow or share sample output.

Which base model (SD 1.5, SDXL, or custom) do you normally train on, and why?

How would you ensure visual consistency of the same character across multiple videos?

What’s your estimated turnaround time to deliver the trained LoRA, workflow setup, and first test clips?