Boost Marketing Design with AI: Stable Diffusion & Runway GEN‑2 Case Studies

With the rise of AIGC, using AI to assist designers and improve efficiency has become a clear trend. This article shares several practical AI‑tool case studies, including the thinking process and exploration steps, to inspire readers.

#01 Stable Diffusion Boosts Marketing Campaign

Stable Diffusion, an AI‑driven painting software, enables designers to create 2D sketches that are efficiently converted into 3D effects, saving considerable drawing time. In the "智行" card activity, illustrations were transformed into a 3D style using Stable Diffusion, reducing production time.

#02 AI‑Generated Dynamic Posters

Runway's GEN‑2 can generate videos from text or images. By feeding descriptive prompts, designers can create coherent, realistic dynamic posters for tourism marketing, enhancing immersion.

Steps:

Split the poster into copy, main character, and background.

Analyze the scene and set a background‑animation prompt such as "a yurt in the desert with lights and a sky full of stars, the stars are twinkling, in the style of hyper‑realistic".

Use the image‑plus‑text mode to generate the video.

#03 LoRA Model Training for Free 3D Travel Backgrounds

Creating 3D travel backgrounds is a frequent but time‑consuming task in marketing design. A custom LoRA style model was trained to generate these backgrounds quickly and at low cost.

Training workflow (six steps):

Collect a high‑quality, style‑consistent 3D travel image dataset.

Clean the dataset, removing noisy elements and standardising formats to JPG/PNG.

Generate base keywords using Stable Diffusion reverse‑prompting, then manually refine tags.

Set training parameters: model.ckpt, resolution 768×1152, 20 epochs, 36 000 steps.

Test the model with XYZ‑axis orthographic renders and compare results across weights.

Iterate based on feedback, refining the dataset and tags until the LoRA model is stable and generalises well.

#04 AI‑Custom Avatar Generation

This section demonstrates how to use Stable Diffusion to create personalized avatar styles, offering a quick way to generate diverse portrait images.

Key points:

Select a model (e.g., anything‑v5‑PrtRE for fresh comic style, toonyou_beta6 for energetic comic, sdxlNijiSpecial_sdxlNijiSE for Disney, sdvn7Nijistylexl_v1 for elegant youthful look).

Use a positive prompt template such as "3d rendering, C4D, best quality, high quality, asian girl" combined with reverse‑engineered tags like "1girl, solo, braid, glasses, ...".

Apply ControlNet with Depth or Lineart preprocessors for better structure.

These cases illustrate flexible use of AI tools to improve workflow efficiency and spark creative ideas. Future updates will continue to explore AI applications in design.