Run Gemma 4 UNCENSORED in ComfyUI (No More Limits!)
Transcript
0:00 · Let's be real. [music] Open-source text models like when three six and Gemma four are absolutely incredible [music] until they hit you with that dreaded I cannot fulfill this request. If you're in the AI influencer business, [music] censorship is the ultimate creativity killer. But what if I told you there's a way to run a completely uncensored version of Gemma four right inside comfy UI? A model that respects your VRAM, analyzes your reference images [music] like a pro, and writes perfect prompts for you, saving you hours of headache.
0:33 · Today, I'm going to show you exactly how to set this up. If you get value out of this, all I ask is that you hit that like button.
0:40 · [music] First things first, fire up comfy UI. If it's been a minute since you last used it, hit up the manager and update your app so we don't get any missing node errors.
0:52 · [music] Don't restart just yet. Head over to the custom node manager, search for comfy UI Ollama, and click install. [music] Once that's done, go ahead and restart comfy UI.
1:03 · [music] Next up, you need Ollama. If you don't have it installed, just go to their official website, grab the download for your operating [music] system, and install it. It's seriously that simple.
1:15 · [music] Now, we aren't using the boring vanilla Google Gemma four. We're using a special version by Fredrizzones55. [music] Huge shout-out to him for this. This is the aggressive version. That means it [music] keeps the original brain and personality of the model, but completely strips away the censorship.
1:36 · It will never refuse a prompt. We're going to grab the quantized Q4KM version. Why? Because it's super lightweight, totally VRAM friendly, and still packs a massive punch. [music] To download it, just copy the run command from the Ollama model page, open up PowerShell, and paste it in. Make sure your Ollama app is running first. I've already got mine downloaded and as you can see in PowerShell, it's ready to [music] rock.
2:04 · All right, close PowerShell and jump back into ComfyUI. Let's drop in an Ollama chat node. [music] Drag out the connectivity input and load the Ollama connectivity node. This is the [music] bridge between your Ollama server and ComfyUI. Make sure [music] Ollama is running in the background.
2:28 · You can now load your Gemma 4 Uncensored model. Here is a massive pro tip. If your GPU is crying for VRAM, set the keep alive value to 0 minutes. This unloads the language model from your VRAM the second it finishes generating text, leaving all the power for your image generation models. Let's hook up a preview text node to see the magic happen. Since Gemma 4 is a vision model, we can plug in an image.
3:01 · And it will describe it in crazy detail. But, a raw description is usually way too long for image generators. [music] That's exactly why I crafted a custom system prompt. When you drop this into the Ollama chat node, it cleans up the description and gives you a flawless optimized prompt ready for models like Flux 2 or Z image. Just look at this clean output. Now, let's bring in a basic Z image Turbo template.
3:39 · We'll connect the text output from our Ollama chat straight into the positive prompt input of our generation node. [music] Now, if you are completely new to this, like downloading checkpoints, clip models, and VAEs that match your graphics card. Don't worry, I have a complete step-by-step masterclass over on my Patreon.
3:59 · [music] In fact, if you want to master the AI influencer game, build your own consistent [music] characters, and actually start making money from your art, my Patreon is the place to be. [music] Click the link in the description to join our community and take your skills to the next level. All right, let's run the workflow. Gemma 4 analyzes our reference image, crafts the perfect prompt, and sends it to Z image turbo.
4:25 · Look at this result. Every detail from the reference is preserved [music] beautifully. The face is a bit different, but that's totally normal since we aren't using character LoRAs right now, but here is where it gets incredibly fun. [music] We left the user prompt box empty, right? Let's use it to our advantage.
4:42 · [music] You can type something simple like, "Preserve the character but change her hair to long and dark." Hit generate, and wow, the model swaps the short red hair for long dark hair perfectly. [music] Let's try a goofy mirror selfie reference image. Sometimes the AI misses the type of shot based just on the system [music] prompt. No problem, just tell the user prompt to make this a mirror [music] selfie.
5:13 · Boom, nailed it.
5:18 · Want a regular selfie instead? Just type selfie. [music] You can change body types, clothing, the background, anything you want.
5:28 · [music] You have absolute creative control to customize the output with your own vision. That is how you unlock the true power of Gemma 4 uncensored [music] inside ComfyUI. Oh, and because this model has zero filters, [music] yes, you can generate highly detailed NSFW photos. I've crafted [music] a very specific system prompt just for that and it's waiting for you exclusively on my Patreon.
5:54 · [music] Thank you so much for watching. If this video helped you, smash that subscribe button so you don't miss out and I'll see you in the next comfy UI tutorial.