4B / 9B / Base vs Qwen 2511 Multi-Image Comparison Workflow

4B / 9B / Base vs Qwen 2511 Multi-Image Comparison Workflow is a ComfyUI comparison workflow designed for testing multi-image editing behavior across several modern image models. It compares FLUX.2 Klein 4B, FLUX.2 Klein 9B, FLUX.2 Klein Base, and Qwen Image Edit 2511 Lightning inside one practical workflow, allowing creators to evaluate how different models handle the same reference images, prompts, logo transfer, style transfer, product editing, object recoloring, and multi-image instruction-following tasks.

This workflow is not a simple text-to-image graph. It is built as a multi-reference image comparison system. The main goal is to help users test how different model families respond when they are given multiple input images and a specific editing instruction. For example, one image can provide the object or product structure, while another image provides the logo, color palette, pattern, material, or style reference. The workflow then lets different models attempt the same transformation so users can compare prompt adherence, reference-image understanding, structural preservation, visual quality, and editing stability.

The FLUX.2 Klein 9B branch uses the flux-2-klein-9b-fp8 model with qwen_3_8b_fp8mixed as the text encoder and flux2-vae as the VAE. This branch is useful for testing stronger instruction following, better multi-image reasoning, and more robust reference transfer. Because the 9B model has more capacity than the 4B variant, it is suitable for tasks that require more precise visual understanding, such as transferring a logo from one reference image onto a product, changing a product color while preserving shape, or maintaining the original lighting and perspective while applying a new visual identity.

The FLUX.2 Klein 4B branch uses the flux-2-klein-4b-fp8 model with a lighter Qwen text encoder setup. This branch is useful for testing speed, lower VRAM pressure, and practical online deployment performance. It may be faster and easier to run than the 9B version, making it suitable for creators who want a more lightweight multi-image editing workflow. Comparing it directly against 9B helps users decide whether the quality gain of the larger model is worth the extra resource cost.

The FLUX.2 Klein Base branch is designed as another reference point in the comparison. It gives users a way to test the base model behavior against the tuned 4B and 9B routes. This is useful for observing differences in visual consistency, reference transfer strength, image cleanliness, object preservation, and prompt responsiveness. In production testing, the base model can sometimes behave more conservatively, while tuned or larger variants may follow complex edit instructions more strongly.

The Qwen Image Edit 2511 branch uses Qwen Image Edit 2511 Lightning LoRA routes, including fast 4-step and 8-step style editing logic. This branch is useful for testing how Qwen handles direct image-editing tasks compared with the FLUX.2 Klein family. Qwen Image Edit 2511 is especially valuable when the instruction is explicit, such as “apply this logo,” “change this color,” “keep everything else unchanged,” or “transfer the style from image 2 to image 1.” By placing it in the same comparison workflow, users can quickly see whether Qwen or FLUX.2 Klein performs better for a specific multi-image edit.

A key feature of this workflow is the ReferenceLatent structure. Reference images are encoded into latent space and then passed into the model conditioning chain. This makes the workflow suitable for multi-image reference editing rather than pure prompt-only generation. The user can provide multiple images: one as the base target and another as the style, logo, color, or design reference. The workflow then compares how each model understands the relationship between those references.

The workflow also includes clear operational sections such as model loading, prompt input, reference image input, image size control, custom sampler, and output saving. This structure makes it easier for users to test several models with the same prompt and source images. Instead of rebuilding separate workflows for each model, this graph keeps the comparison environment more consistent.

This workflow is especially useful for product image editing. For example, you can take a handbag image and apply colors or logos from a second reference image. You can take a car interior image and apply a logo to the steering wheel while keeping the original camera angle, lighting, material, and surrounding details unchanged. You can test product recoloring, branding replacement, package design transfer, clothing style transfer, and multi-reference visual adaptation.

It is also useful for model evaluation. When comparing AI models, a single pretty output is not enough. A good comparison should test instruction accuracy, reference consistency, object structure, texture quality, edge blending, text/logo clarity, color transfer, identity preservation, and artifact control. This workflow is designed exactly for that kind of practical comparison.

Main features:

- Multi-model comparison workflow for ComfyUI

- FLUX.2 Klein 9B FP8 branch

- FLUX.2 Klein 4B FP8 branch

- FLUX.2 Klein Base branch

- Qwen Image Edit 2511 Lightning branch

- Multi-reference image editing test

- ReferenceLatent-based image conditioning

- Product logo transfer testing

- Color and style transfer testing

- Object-preserving image editing

- Same-prompt comparison across different models

- Flux2Scheduler and SamplerCustomAdvanced support

- VAEEncode and latent reference workflow

- SaveImage output for model-by-model comparison

- Useful for RunningHub online testing and Civitai showcase examples

Recommended use cases:

Multi-image editing tests, model comparison, FLUX.2 Klein 4B testing, FLUX.2 Klein 9B testing, FLUX.2 Klein Base testing, Qwen Image Edit 2511 testing, product image editing, logo transfer, color transfer, brand design adaptation, handbag customization, car interior logo editing, package design testing, reference image transfer, commercial visual testing, ComfyUI workflow evaluation, RunningHub workflow publishing, and Civitai comparison posts.

Suggested workflow:

Start by preparing the reference images. For best results, use one clear target image and one clear style or logo reference image. The target image should contain the object you want to edit, such as a handbag, steering wheel, product package, clothing item, or interior object. The reference image should clearly show the color, logo, texture, design element, or style you want to transfer.

Then write a direct editing prompt. For multi-image editing, the prompt should clearly explain the relationship between the images. For example: “stylize the handbag in image 1 with the colours and logo from image 2,” or “apply the yellow C logo to the center hub of the steering wheel, and change the steering wheel color to royal blue matching the logo background, while maintaining the same interior style, lighting, camera angle, and all other elements unchanged.” This kind of prompt is more useful than a vague artistic description because it tells the model exactly what to preserve and what to change.

Keep the comparison controlled. Use the same prompt, the same reference images, and similar output size when comparing 4B, 9B, Base, and Qwen 2511. If every model receives different inputs, the comparison becomes less meaningful. The main value of this workflow is that it allows model differences to become visible under similar conditions.

Pay attention to prompt adherence. A model may produce a beautiful image but still fail the task if it does not apply the correct logo, changes the wrong object, destroys the original product structure, or ignores the reference image. When evaluating results, check whether the model followed the instruction first, then judge visual quality.

For logo transfer, check logo placement, shape accuracy, readability, edge clarity, and whether the logo blends naturally with the surface. For color transfer, check whether the color matches the reference without destroying material realism. For product editing, check whether the object shape, camera angle, lighting, and background remain stable. For style transfer, check whether the model transfers the intended style without over-changing the whole image.

The 9B branch is usually the better candidate for more complex reference understanding, but it may require more resources. The 4B branch is useful for faster testing and lighter deployment. The Base branch helps evaluate the underlying model behavior. The Qwen 2511 branch gives a strong editing-oriented comparison point, especially for direct instruction-based image edits.

For practical testing, run a small number of cases first. Start with a simple logo transfer or product recolor. Then test harder tasks such as multi-object editing, detailed brand replacement, material transfer, or keeping a complex background unchanged. Save each model’s output separately and compare them side by side.

This workflow is designed for creators who want to evaluate real multi-image editing ability rather than only text-to-image beauty. It helps identify which model is better for commercial editing, product mockups, logo transfer, style transfer, and reference-based image generation. By comparing FLUX.2 Klein 4B, FLUX.2 Klein 9B, FLUX.2 Klein Base, and Qwen Image Edit 2511 in one workflow, users can make more informed decisions before building a production pipeline.

🎥 YouTube Video Tutorial

Want to know what this workflow actually does and how to start fast?

This video explains what the tool is, how to launch the workflow instantly, and shares my core design logic — no local setup, no complicated environment.

Everything starts directly on RunningHub, so you can experience it in action first.

👉 YouTube Tutorial: https://youtu.be/7byGk8b04Ao

Before you begin, I recommend watching the video thoroughly — getting the full context helps you understand the tool faster and avoid common detours.

⚙️ RunningHub Workflow

Try the workflow online right now — no installation required.

👉 Workflow: https://www.runninghub.ai/post/2013130226275983361?inviteCode=rh-v1111

If the results meet your expectations, you can later deploy it locally for customization.

🎁 Fan Benefits: Register to get 1000 points + daily login 100 points — enjoy 4090 performance and 48 GB super power!

📺 Bilibili Updates (Mainland China & Asia-Pacific)

If you’re in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown.

📺 Bilibili Video: https://www.bilibili.com/video/BV1EKkuBbE4R/

☕ Support Me on Ko-fi

If you find my content helpful and want to support future creations, you can buy me a coffee ☕.

Every bit of support helps me keep creating — just like a spark that can ignite a blazing flame.

👉 Ko-fi: https://ko-fi.com/aiksk

💼 Business Contact

For collaboration or inquiries, please contact aiksk95 on WeChat.

🎥 YouTube 视频教程

想了解这个工作流到底是怎样的工具，以及如何快速启动？

视频主要介绍 工具定位、快速启动方法 和 我的构筑思路。

我们会直接在 RunningHub 上进行演示，让你第一时间看到实际效果。

👉 YouTube 教程： https://youtu.be/7byGk8b04Ao

开始前建议尽量完整地观看视频 —— 把握整体思路会更快上手，也能少走常见弯路。

⚙️ 在线体验工作流

现在就可以在线体验，无需安装。

👉 工作流： https://www.runninghub.ai/post/2013130226275983361?inviteCode=rh-v1111

打开上方链接即可直接运行该工作流，实时查看生成效果。

如果觉得效果理想，你也可以在本地进行自定义部署。

🎁 粉丝福利： 注册即送 1000 积分，每日登录 100 积分，畅玩 4090 体验 48 G 超级性能！

📺 Bilibili 更新（中国大陆及南亚太地区）

如果你在中国大陆或南亚太地区，可以通过下方视频查看该工作流的实测效果与构思讲解。

📺 B站视频： https://www.bilibili.com/video/BV1EKkuBbE4R/

我会在 夸克网盘 持续更新模型资源：

👉 https://pan.quark.cn/s/20c6f6f8d87b

这些资源主要面向本地用户，方便进行创作与学习。