Over the last couple of months a colleague and I working together at Versus have been developing a bespoke node-based AI generative toolkit. ( like Krea, or Flora ). The API calls are pennies on the dollar from any 3rd party tool, and the UI is bespoke to exactly like how we want to work.
That phrase, vibe coding, gets thrown around a lot right now, but in this case it is exactly what the process felt like. I was not writing a formal spec from the top down. I was working from friction. I would hit something annoying in the creative workflow, describe what I wanted the tool to feel like, test it, react to it, and then keep shaping it.
The app evolved through conversation, screenshots, broken runs, UI instincts, and a lot of small decisions that slowly turned into a real system.
Newt_Node started as a local node-based workflow for image and video generation. It has now become the center of how I am doing AI visual development. At this point, it has effectively replaced my use of Krea and Flora for this kind of work.
Not because those tools are bad. They are polished, capable, and useful. But they are also someone else's workflow.
Newt_Node is bespoke, and I'm in full control of everything.
Why Replace Krea And Flora?
The biggest shift is control.
In hosted creative tools, I am usually working inside a beautifully designed box. That can be great when the box matches what I want to do. But when I want to build a repeatable workflow, inspect costs, reuse intermediate outputs, route prompts through different processors, connect images into video models, or keep a local project structure, I start wanting something more modular.
Newt_Node gives me a graph instead of a single prompt box.
The graph matters because creative AI work is rarely just one prompt anymore. It is a text idea, a prompt processor, a reference image, a style reference, a composition guide, a video source, a mask, a utility pass, a preview, a generation, and another generation based on the first result.
Text nodes are LLMs and can be setup as agents with specific tasks. It's all so crazy to me, I love it.
Building The Tool While Using The Tool
The funny thing about Newt_Node is that it was built in the same spirit that it supports.
I would use the app, find the next missing behavior, and then add it. The canvas needed to pan with a regular mouse, not just a touchpad. The canvas needed to resize with the browser window. Save and Open needed to preserve projects. Previews needed to hold onto multiple generations. Stats needed to show real cost estimates, because guessing at spend is a terrible way to use paid generation APIs.
Each improvement came from a real moment of use.
The app grew node by node: Text nodes for prompt processing, Image Model nodes for generation, Video Model nodes for motion, Utility nodes for cleanup and transformation, Preview nodes for comparing outputs, Groups for organizing workflows, Composer for shot blocking, and now a dedicated 3D node for asset generation.
That last one is especially interesting because it starts to pull Newt_Node beyond image and video prompting and into asset creation.
The Composer Node
One of the more important additions has been the Composer node.
The idea is simple: I want a lightweight way to block a shot before I ask an image model to generate it. Not a full 3D package. Not Blender. Just a fast composition space with a camera, human maquettes, primitives, image planes, and a frame output.
That frame can then be connected into an image model as a composition guide.
This matters because image models are strong at style and detail, but they do not always respect layout unless the instruction is very clear. A rough 3D frame gives the model a visual anchor. The better the maquette and camera framing, the easier it is to push composition instead of just hoping the prompt lands.
The Composer node is not trying to be final art. It is trying to be visual intent.
The Image-to-3D Node
The latest addition is a dedicated 3D node using Hunyuan 3D through Fal.
When the model runs, Newt_Node downloads the generated GLB locally, stores it in the outputs folder, previews it inside the node with a Three.js viewer, and provides a download button directly on the result. Generates a 'photogrammetry geo + textures.
That may sound like a small thing, but it changes the feel of the workflow. The result is not trapped in a provider page. It is part of the project graph. It can be previewed, saved, downloaded, and used as a step in a larger creative process.
Utility Nodes And Video Workflows
Another area where Newt_Node has grown quickly is the Utility node.
Video cleanup is rarely one model and done. A shot might need a mask, a matte, interpolation, upscaling, frame extraction, or inpainting. Having those operations live in the same canvas as the creative generation nodes makes the work easier to understand and repeat.
The important part is that every pass stays visible. I can see what source video went into the node, what model was used, where the mask came from, and where the result went next.
That visual continuity is one of the reasons I keep reaching for Newt_Node first.
I'm sure many of these 'features' are capable of being prompted with Nano Banana, GPT image 2. However, I've added complete control over those generations. No longer rolling the dice and hoping I get back what I wanted.
Why This Feels Different
The reason Newt_Node has replaced Krea and Flora for me is not just that it can call similar models. It is that it reflects the way I actually want to work, and there is no 3rd party markups or function limitations. Works exactly how I do. It's a bespoke app!
I want to connect ideas visually, keep prompts and outputs together, use images, video, text, masks, and 3D as equal citizens, build workflows that can be saved and reopened, see which models are active, inspect and reuse generations, understand cost, and add new model types without redesigning the whole app.
That is the real win.
Newt_Node is not just a wrapper around generation APIs. It is becoming a personal creative operating system for AI media work.
Codex: Vibe Coding As A Collaboration
The development process itself has been a collaboration between instinct and implementation.
I would say things like: this node feels too cluttered, the dots should move when Settings opens, the preview should not clear after a new generation, I need to delete selected connection lines, I want grouped nodes to move together, I need the video preview to loop, I want Run All to play selected preview videos, I want the 3D input dots broken out by view.
Those are not abstract feature tickets. They are creative workflow observations.
The app got better because the feedback was grounded in actual use. Screenshot by screenshot, run by run, annoyance by annoyance.
That is the part of vibe coding I actually like. It is not about being careless. It is about staying close to the experience of the tool while building it.
Where It Goes Next
The next step is production.
These tools are not just interesting experiments anymore. They are going to be used to create imagery and video in ways, and at speeds, that were impossible before. The important part is not only that the machine can generate more. It is that I can bring 20+ years of creative direction to the process and curate the slop into something intentional.
That is where this starts to feel empowering.
Newt_Node gives me a way to control media instead of just prompting and hoping. I can build a workflow, steer the references, shape the composition, review the outputs, track the cost, and keep the whole process inside a sandbox that is tuned to how I actually think.
I am already cooking up partnerships for media creation, and I can see how this will blow commercial clients away with visuals that move faster from idea to execution than the old pipeline could allow.
Yes, a lot of this is possible with third-party apps. Krea and Flora are still out there, and they are still powerful. But there is something incredible about having a bespoke sandbox where I can roll my own.
That is what Newt_Node is becoming: not just another AI interface, but a production environment for making the work sharper, faster, stranger, and more mine.
Comments