I recently spent time testing local LLMs and image generation workflows using apps like ComfyUI and InvokeAI on Windows. I wanted to see just how far I could push a local setup with the RTX 4070.
The Experience
All in all, the easiest way to go about image generation without restrictions is using ComfyUI. It offers the most control. While there are simpler alternatives, ComfyUI is the most robust option for this specific rig.
Video Generation
Video generation is where the dream starts to crack. Unless you have a super PC, I would suggest forgetting about it. It simply isn’t worth it given how long it takes and how random the output can be.
It is still nothing I would utilize in a real project. The quality is generally low, and the clips are very short. It seems easier to just buy footage if needed rather than making my own AI-generated content. It is not worth the processing power to create 2 seconds of video, especially when you don’t even know what will be created until the render finishes.
The Power of the 4070 vs. Time
With the RTX 4070, you technically have the power to:
- Run 70B LLMs locally
- Run SDXL at full res
- Run Flux models smoothly
- Run Stable Video Diffusion for multi-second videos
- Run multiple models at once
- Run ControlNet + upscaling + refiner
- Do full AI pipelines (image → video → upscale → VAE decode)
However, just because you can do these things doesn’t mean it is efficient. The timing is what makes the process not worth it for many of these tasks. While the card can handle the load, the wait times for complex pipelines—especially video—can be prohibitive, making it less practical than one might hope.
Cost Comparison: Local vs. Cloud
When looking at the economics, the trade-off is clear.
Local (RTX 4070):
- Upfront Cost: High (GPU + System).
- Recurring Cost: Electricity.
- Privacy: 100% private.
- Freedom: No censorship or guardrails.
Cloud (Midjourney/OpenAI):
- Upfront Cost: $0.
- Recurring Cost: $20-$60/month subscriptions.
- Speed: Generally much faster.
- Quality: Often higher consistency with less tweaking.
If you are just tinkering, the local setup pays for itself over time by avoiding subscriptions. However, if you need professional results fast, the cloud services are often worth the monthly fee just to save hours of debugging and rendering time.
Blog