# Flux.1 Control Net

**Testing Setup**:

* Base Model: `black-forest-labs/FLUX.1-dev`
    
* LoRA Model: `prithivMLmods/Retro-Pixel-Flux-LoRA` (scale=0.8)
    
* ControlNet: `Shakker-Labs/FLUX.1-dev-ControlNet-Union-Pro`
    
* Prompt: `8 by 8 Retro Pixel Art Character Sheet of NPC characters for pixel art game called Machi.`
    

**Tile Control Net Image:**

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1750977210001/7d007fa2-7bf2-46e9-b420-f22c3c6ee000.png align="center")

**Without control net:**

**With** `controlnet_conditioning_scale` **of 0.25:**

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1750977289517/003321bf-c9f6-49fb-9953-7d1303adaf7c.jpeg align="center")

**With** `controlnet_conditioning_scale` **of 0.30:**

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1750977333228/3adb120a-e199-4c73-93da-b15b1361f4e8.jpeg align="center")

From what I can tell, the control net is able to help with the **layout**.

**With** `controlnet_conditioning_scale` **of 0.50:**

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751025262834/5b51bf7b-95a5-4ffb-8dc6-9d86702a995f.jpeg align="center")

Somewhere between 0.3 and 0.5, the control net becomes too strong. At that level, the image just becomes a mere copy with some dittering.

**With** `controlnet_conditioning_scale` **of 0.35:**

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751025525583/b640ac61-db59-4c62-9b1b-b06ec272f48c.jpeg align="center")

0.35 is still strong.

Now, let's keep the `controlnet_conditioning_scale` at **0.30**, and variate the template image resolution. Above ones were done with 1024×1024 templates.

**128×128 Template Image:**

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751026693057/ec419a66-d23b-4888-8897-9938d9b1f7da.png align="center")

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751026744325/17eae9d5-04b4-447f-bdbd-28e778d442a0.jpeg align="center")

**64×64 Template Image**:

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751025744424/df700b7a-cd09-498c-82fd-0291cc2b6cfa.png align="center")

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751025828593/68d38f9a-77e7-4e42-8b94-c453b4aa7093.jpeg align="center")

64×64 still works on providing the overall layout. There doesn’t seem to be much difference from 128×128.

**32×32 Template Image**:

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751026210390/5ea85abf-0af2-4544-b00c-b0fb785d1de2.png align="center")

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751026224060/baf9a603-786a-40cb-b37a-57d6aec4185a.jpeg align="center")

At this resolution, the layout is still preserved.

**16×16 Template Image**:

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751026371506/bd16e962-cb92-4eaa-93a5-20757ee32c31.png align="center")

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751026461856/03b01585-9785-4b4f-8089-b722ad7016c6.jpeg align="center")

At 16×16, the grid collapses and does not give much meaning.

---

Now, let’s fix the template resolution to **1024×1024** and variate zoom with nearest neighbor sampling.

* `lora_scale` still fixed at **0.8**.
    
* `controlnet_scale` still fixed at **0.3**.
    

**2x Zoom Template:**

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751027676544/659a299f-bd8f-4d4a-8ba4-9fe61410ed69.png align="center")

> 4 by 4 Retro Pixel Art Character Sheet of NPC characters for pixel art game called Machi.

**Generated:**

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751027730538/e3d882a6-d2e4-4e3d-adb1-1ffe24e5ee2f.jpeg align="center")

**4x Zoom Template:**

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751027965251/b5059aeb-dd43-4012-8cce-5027b15bb41d.png align="center")

> 2 by 2 Retro Pixel Art Character Sheet of NPC characters for pixel art game called Machi.

**Generated:**

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751028010218/9b45aff8-b7b7-486c-ade7-4b5fd786763e.jpeg align="center")

When using nearest neighbor downsampling, we get the following:

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1751028511475/c38777ee-d663-4aae-8d66-4661bd2cf833.png align="center")

It does become blurry. We probably can find a better downsampling algorithm…

But, so far from the past generations, 4x zoom seems to be most stable.

---

To be continued…