Back to home

Hy Image 3.5 in Miora: A Design Evaluation Guide

Understand Hy Image 3.5 in Miora and plan useful image generation, typography, reference fidelity and editing checks.

Last updated: 2026-10-06

Separate the model from the workspace

The official Miora site lists Hy Image 3.5 Preview among its image model offerings. Miora is the creative workspace; Hy Image 3.5 is a model used within it. Account access, interface controls and product limits are separate from the capabilities an API may expose.

Explore the Miora Design workspace for the creation tools available on this site. This page does not assert a reference-image limit, maximum resolution, price or free-access deadline.

Define a useful image task

Choose a task you can inspect: a product campaign with a fixed reference, a poster with exact text or an edit that changes only one element. Write the checklist before generating.

For a product image, check shape, label, material, color and small construction details. For a poster, read all text and inspect the hierarchy. For an edit, compare both the requested change and everything that should remain unchanged.

Record the conditions

Keep the platform, model name, date, prompt, references, aspect ratio, settings and number of attempts. A comparison is easier to interpret when these conditions are visible.

If you compare models, use the same brief, reference material and attempt budget. Show failed outputs as well as successful ones, and distinguish observation from explanation.

What this guide does and does not claim

This is an evaluation method, not a completed benchmark. We do not publish invented scores, claim superiority over another model or attribute this template's images to Hy Image 3.5. Apply the checklist to your own results before deciding whether a workflow is ready for production.

A small evaluation you can repeat

Use three clearly defined tasks rather than judging a workflow from one attractive image. The following is a proposed test plan, not a completed benchmark:

  1. Reference preservation: ask for a new background around one product. Compare the silhouette, cap, label position, color and material with the reference.
  2. Text handling: request a short approved headline. Transcribe the visible output and compare every character with the brief. Check reading order separately from spelling.
  3. Controlled editing: change only the background color. Compare the product, lighting, shadows and crop before and after the edit.

Decide the attempt budget before starting. Keep all outputs, including rejected ones. If you change the reference or prompt halfway through, record it as a different test condition rather than silently combining the results.

Example prompt for a controlled edit

Edit the uploaded product image by changing only the background to pale blue.
Preserve the product's shape, material, color, label and position.
Keep the camera angle, crop, light direction and existing product shadows.
Do not add text, props or another product.

This illustrative prompt makes the requested change inspectable. If the product or crop also changes, record that as an unintended edit even when the final image looks appealing.

Keep an observation log

For each attempt, record the task, exact prompt, reference files, displayed model name, date and output. Describe concrete observations such as “label moved to the left” or “background changed but bottle color also shifted.” Mark the asset accepted only when it passes the requirements that matter for the destination. Without your own recorded outputs, this guide cannot establish how well a model performs on your products.

Continue exploring

Miora AI: A Practical Guide to Miora Design

Miora Design Tutorial: From Brief to Reviewed Visual

Miora Design Prompts: Product, Poster & Storyboard Frameworks

Questions or feedback

Use the support entry on our homepage to send feedback about this guide. Miora Design Lab is independent of Tencent and Miora.