There are hundreds of AI creative tools. Most are wrappers around the same models. A few are genuinely different. This is a guide to the ones that matter — the platforms and tools that working creatives actually build their pipelines around.

The three approaches

Every AI creative workflow falls into one of three categories:

Direct model access. You use a model’s own platform — Midjourney’s website, Google AI Studio for Nano Banana, ChatGPT for GPT Image. Simple. One model, one interface, one subscription. Best when you know exactly which model you want and the job is straightforward.

Multi-model platforms. You access many models from a single workspace. Flora Fauna, Weavy, and similar tools let you switch between Nano Banana, Flux, GPT Image, Kling, Veo, and others without leaving the canvas. Best when your workflow involves multiple models or you want to compare outputs.

Self-hosted pipelines. You run models locally on your own hardware. ComfyUI is the dominant tool here. No per-image cost, total control, full privacy. Best when you need customisation, fine-tuning, or high-volume output without recurring costs.

Most working creatives end up using a combination. Direct access for quick one-offs, a multi-model platform for production work, and ComfyUI for anything that needs fine-tuned models or specialised pipelines.

Multi-model platforms

Flora Fauna

A node-based infinite canvas with 80+ AI models — image, video, text, upscaling, editing. You connect models visually, building pipelines where the output of one model feeds into the next. An image generation → upscaling → video animation pipeline is a single canvas workflow.

Flora launched FAUNA on April 3, 2026 — an AI agent that builds these workflows from natural language. You describe what you want and FAUNA selects the models, writes the prompts, and sequences the steps. Three modes: Assist (asks before running), Auto (runs immediately), Plan (ideation only, no cost).

Models available: Nano Banana, Flux, GPT Image, Stable Diffusion, Ideogram, Recraft, Veo 3.1, Kling, Seedance, Runway Gen-4.5, Pika, and many more. Not Midjourney (no public API).

Pricing: Free tier (1,000 credits), Starter $18/month (20,000 credits), Studio $54/month, Scale $200/month. All plans include all models. No per-seat charges.

Best for: Multi-step creative workflows, teams, comparing models on the same brief, production pipelines that chain image → upscale → video.

Runway

Runway has moved beyond pure generation into a creative production suite. Gen-4.5 is their current video model — strong on visual fidelity and temporal consistency.

Runway’s position in 2026 is increasingly about concept development, previsualisation, and rapid creative experimentation. Production companies use it for testing ideas before committing to full shoots.

Best for: Video-first workflows, filmmakers, ad teams, rapid visual prototyping.

Weavy

A newer entrant focused on creative workflow automation. Less model breadth than Flora but streamlined for specific use cases like marketing asset production.

Best for: Marketing teams, campaign asset generation at scale.

Self-hosted: ComfyUI

ComfyUI is the most powerful and the most intimidating tool in this list. It is a node-based visual interface for building generative AI workflows that run locally on your hardware.

In 2026, ComfyUI has evolved significantly:

App Mode (new in 2026) transforms complex node workflows into simple interfaces. You build the pipeline once, then expose only the controls that matter — prompts, sliders, image inputs — in a clean UI that hides the underlying graph. Non-technical team members can run your workflows without understanding nodes.

ComfyHub lets you share workflows as distributable applications. Build a workflow, publish it, and anyone can run it.

Nodes 2.0 replaced the legacy rendering engine with a modern Vue.js architecture. Over 1,000 custom node packages are available. The ecosystem includes inpainting, LoRA integration, ControlNet, IP-Adapter, and model-specific optimisations.

The trade-off: ComfyUI requires a dedicated GPU (minimum 8 GB VRAM for basic work, 16+ GB for comfortable use, 24+ GB for Flux 2 at full quality). Setup involves Python, CUDA drivers, and model downloads. The learning curve is steep, but the ceiling is the highest of any tool on this list.

Why it matters: Zero per-image cost. Total privacy — nothing leaves your machine. And the deepest customisation: fine-tuned models, trained LoRAs, precise control over every generation parameter.

Best for: Technical creatives, custom style training, high-volume production, privacy-sensitive work.

Upscaling

Most AI models generate at 1-1.5K resolution. Print and production work requires more. Upscaling is not optional.

Topaz Gigapixel AI — the industry standard. Desktop app, one-time purchase. 2-6x upscale with AI detail synthesis. Used by print photographers and production studios.

Magnific AI — web-based upscaler that “reimagines” during upscaling. Adds detail and texture. Can enhance AI art beautifully but may change the image character. Best for illustration and stylised work.

Built-in platform upscalers — Flora Fauna integrates both Magnific and Topaz as nodes in its canvas. Midjourney has 2x upscale buttons. These are convenient but often insufficient for large-format print.

Video editing

AI generates clips. You need an editor to make them into anything longer than 8 seconds.

DaVinci Resolve — free and professional-grade. The standard recommendation. Colour grading, audio editing, and visual effects in one app. The learning curve is real but the tool is genuinely world-class.

Adobe Premiere Pro — if you already have a Creative Cloud subscription. Integrating AI-generated clips into existing production workflows.

CapCut — free, simple, good enough for social media. If the output is a 30-second Instagram reel, this is faster than Resolve.

Audio

AI video models increasingly generate audio, but post-production audio is still essential for finished work.

Veo 3.1 generates native audio — dialogue, SFX, ambience — as part of video generation. The strongest native audio of any current model.

Kling 2.6 generates synchronised audio and can synthesise voices from audio samples.

For everything else: Layer additional audio in your video editor. Free sound libraries (Freesound.org, Pixabay) for effects, licensed music libraries (Artlist, Epidemic Sound) for scores.

The stack that works

For a working AI creative in 2026, here is a practical starting stack:

Image generation: Nano Banana 2 (via Google AI Studio, free tier) for everyday work. Midjourney for editorial and conceptual. Ideogram for text-heavy graphics.

Video generation: Veo 3.1 (free tier, 10 clips/month) for experimentation. Kling or Runway for production.

Multi-model workflows: Flora Fauna when you need to chain multiple models or compare outputs.

Self-hosted: ComfyUI when you need custom models, privacy, or high volume.

Upscaling: Topaz Gigapixel for print. Magnific for stylised enhancement.

Editing: DaVinci Resolve for video. Photoshop or Affinity Photo for image post-production.

You do not need all of these on day one. Start with the free tiers — Google AI Studio and Veo via Google Vids give you image and video generation at no cost. For multi-model access, Flora gives you 80+ models on one canvas — 25% off for 12 months. Add tools as your workflow demands them.

Art & Algorithms publishes guides, tutorials, and prompt packs at the intersection of art and code. Subscribe for the full archive.