Generative AI Landscape 2026
Click any tool to open its website · Hover for details · Click any section header to expand or collapse
🌐 Other AI Fields — not covered in this map
⚡ Automation · n8n, Make, Zapier
🤖 AI Agents · autonomous task execution
💻 Vibe Coding · Cursor, Lovable, Windsurf
🔍 Research AI · Perplexity, NotebookLM
🎙️ Real-Time AI · live voice, live video
🧬 Science AI · protein folding, drug discovery
Focus of this map
✨ Generative AI
AI that creates new content — images, video, voice & music
↓ ↓
Closed Source
Developed by companies to monetize their research. You access them through a subscription or website — the model weights are private, you never own or control the AI. Quality is often top-tier, but you depend on their pricing and terms.
🖼️ Image Generation
▾
T2I — Text → Image
I2I — Image → Image
T2I = write a prompt, AI generates the image from scratch.
I2I = upload a reference image → AI edits, transforms, or remixes it. Used for product shots, style transfer, inpainting (removing or replacing parts of an image).
I2I = upload a reference image → AI edits, transforms, or remixes it. Used for product shots, style transfer, inpainting (removing or replacing parts of an image).
🎬 Video Generation
▾
T2V — Text → Video
I2V — Image → Video
V2V — Video → Video
T2V = text prompt → AI renders a clip from scratch.
I2V = upload an image as the first frame → AI animates it into motion.
Start / End Frame = give AI a start image AND an end image → it generates the transition between them. Full control over movement.
Keyframe technique = define frames at specific timestamps → AI fills the motion between each.
I2V = upload an image as the first frame → AI animates it into motion.
Start / End Frame = give AI a start image AND an end image → it generates the transition between them. Full control over movement.
Keyframe technique = define frames at specific timestamps → AI fills the motion between each.
🎙️ Voice & Music
▾
T2A — Text → Voice
Voice Clone
T2M — Text → Music
T2A = write a script → AI speaks it in a realistic voice.
Voice cloning = upload ~30 sec of your voice → AI speaks in your exact voice for any text.
T2M = describe a genre, mood, or paste lyrics → AI generates a full produced song with instruments and vocals.
Voice cloning = upload ~30 sec of your voice → AI speaks in your exact voice for any text.
T2M = describe a genre, mood, or paste lyrics → AI generates a full produced song with instruments and vocals.
Open Source
Model weights are publicly released — anyone can download, self-host, and modify them. Lower running costs at scale, no censorship, full data privacy. Requires more technical setup: run locally on a GPU or rent cloud compute by the hour.
🖼️ Image Generation
▾
T2I — Text → Image
I2I — Image → Image
LoRA Fine-tuning
Open source image models run via ComfyUI (node-based workflow editor) or Automatic1111. Download the model weights once and run them on any GPU — locally or on a cloud rental like RunPod. LoRA = lightweight add-on files that steer the base model toward a specific style, face, or product.
🎬 Video Generation
▾
T2V — Text → Video
I2V — Image → Video
Start/End Frame
Open source video models require a beefy GPU (24GB+ VRAM recommended) or cloud GPU rental. Run via ComfyUI video nodes or dedicated UIs. Generation is slower than closed source but costs fractions of a cent per clip at scale.
🎙️ Voice & TTS
▾
TTS — Text → Voice
Voice Clone
V2V — Voice Conversion
TTS = text to speech, fully open source.
Voice cloning = train on a few seconds of audio → model speaks in that voice.
V2V / Voice Conversion = run any voice through a trained model to make it sound like a specific person. Real-time or batch. This is the open source answer to ElevenLabs' voice changer.
Voice cloning = train on a few seconds of audio → model speaks in that voice.
V2V / Voice Conversion = run any voice through a trained model to make it sound like a specific person. Real-time or batch. This is the open source answer to ElevenLabs' voice changer.
🧬 LoRAs & Custom Checkpoints
▾
LoRA Training
Custom Checkpoint
Style / Face / Product
What is a base model? Think of it as a general artist — it can draw anything, but it doesn't know your face, your product, or your brand style.
What is a LoRA? A LoRA (Low-Rank Adaptation) is a small add-on file — usually 50–300MB — that you train on top of a base model. You give it 10–30 reference images plus short text descriptions. After training, every image gets "steered" toward what you showed it. Train on your face → model generates your face. Train on a product → model generates that product. Train on an art style → every output looks like that style.
What is a custom checkpoint? When you permanently merge a base model with one or more LoRAs into a single file. Easier to share and deploy — no need to load the LoRA separately each time.
Easiest training tool: AI Toolkit by Ostris — open source, runs on a single GPU, simple config file. Best for FLUX LoRA training. Upload images, write captions, run the script, done.
What is a LoRA? A LoRA (Low-Rank Adaptation) is a small add-on file — usually 50–300MB — that you train on top of a base model. You give it 10–30 reference images plus short text descriptions. After training, every image gets "steered" toward what you showed it. Train on your face → model generates your face. Train on a product → model generates that product. Train on an art style → every output looks like that style.
What is a custom checkpoint? When you permanently merge a base model with one or more LoRAs into a single file. Easier to share and deploy — no need to load the LoRA separately each time.
Easiest training tool: AI Toolkit by Ostris — open source, runs on a single GPU, simple config file. Best for FLUX LoRA training. Upload images, write captions, run the script, done.
How to Access These Models
Closed Source Access
Three ways to use closed source AI — direct website, API, or aggregator platforms.
🌐 Direct Website
▾
The simplest option. Go to the model's website (midjourney.com, kling.ai, suno.com etc.), create an account, buy a subscription or credits, and use their web interface. No technical knowledge needed. Best for beginners or low-volume use.
⚙️ API Access
▾
API = Application Programming Interface. In plain English: a way for one piece of software to talk to another. Instead of clicking a button on a website, you send a request — your prompt, settings, and API key — directly to the AI company's servers, and they send back the result.
Why use it? You can automate everything. Hook it into n8n, build your own app, generate 1000 images overnight without clicking anything. You pay per generation instead of a flat monthly subscription — cheaper at scale.
An API key is like a password that identifies your account. Include it with every request. Most closed source tools offer API access: OpenAI, ElevenLabs, Kling, Runway.
Why use it? You can automate everything. Hook it into n8n, build your own app, generate 1000 images overnight without clicking anything. You pay per generation instead of a flat monthly subscription — cheaper at scale.
An API key is like a password that identifies your account. Include it with every request. Most closed source tools offer API access: OpenAI, ElevenLabs, Kling, Runway.
🗂️ Aggregator Platforms
▾
Websites that plug into multiple AI models and give you one subscription to access all of them — with a simple interface, no API setup required. Pay a monthly fee for a pool of credits, spend them across different models. Good middle ground between "just the website" and "setting up your own API."
Open Source Access
Four ways to run open source models — hosted API, your own machine, or rented cloud GPUs.
⚙️ Hosted API
▾
Even open source models can be accessed via API. Third-party providers host the models on their own servers and charge per generation — no GPU needed on your end. Same automation benefits as closed source API, but running open source models.
🖥️ Your Own PC
▾
If your computer has a dedicated GPU you can run open source models locally — completely free after setup, no per-generation cost, full privacy.
Images (FLUX, SD XL): 8–12 GB VRAM minimum, 16 GB+ recommended.
Video (Wan 2.2, LTX): 24 GB+ VRAM recommended — most consumer cards won't handle video.
Interface: ComfyUI (most powerful, node-based) or Automatic1111 (simpler, image-focused). Install once, runs forever.
Images (FLUX, SD XL): 8–12 GB VRAM minimum, 16 GB+ recommended.
Video (Wan 2.2, LTX): 24 GB+ VRAM recommended — most consumer cards won't handle video.
Interface: ComfyUI (most powerful, node-based) or Automatic1111 (simpler, image-focused). Install once, runs forever.
☁️ Rented Cloud GPUs
▾
Rent a powerful GPU server by the hour — access via browser, install your models, generate, then shut it down. You only pay while it's running. Best of both worlds: open source freedom at professional GPU power, without buying hardware.
Where to Find Models & LoRAs
📦 Model & LoRA Repositories
▾
These are the three platforms you will use constantly to find, download, and share AI models, LoRAs, and checkpoints. Bookmark all three — they are the foundation of the open source AI ecosystem.
🐙 GitHub
Where new models are first released by research labs. Download weights, read the code, follow updates. Start here for any new model.
🤗 Hugging Face
The central hub for downloading model weights and LoRA files. Hosts nearly every major open source model. Best for base models and datasets.
🗃️ Civitai
Biggest library of community-trained LoRAs and checkpoints. Search before training — someone may have already made what you need. Best for image LoRAs and styles.