Video models
Wan 2.2 5B vs 14B: which Wan to download for your graphics card
Wan 2.2 ships as two different models: the 5B single checkpoint that makes 720p clips on one consumer card, and the two-expert 14B pair that gives the best open video quality but wants 24 GB or more. What each one is, how much memory they really take, which LoRAs work with which, how long a 5-second clip takes, and the short answer for 8, 12, 16, 24 and 32 GB cards.
Errors & fixes
How much VRAM do I actually need for local AI?
The honest answer, by model rather than by budget: 8 GB runs 8B chat models and SDXL, 12 to 16 GB runs 14B models, Flux fp8 and short video, 24 GB is the sweet spot for 30B-class LLMs and FLUX.2 or Qwen Image, 48 GB and up is 70B territory, and 64 GB of system RAM is not a 64 GB GPU. With the arithmetic for weights, context and headroom, a memory tier table from 8 GB to 512 GB, and Kai's video on every way to reach 128 GB of VRAM.
Errors & fixes
Safetensors vs .ckpt and .pt: how to download AI models safely and verify them with SHA-256
Why .ckpt and .pt files can run code when loaded and .safetensors cannot, how to recognise a converted or renamed file, how to check a download against the SHA-256 and AutoV2 hashes shown on the model pages, where to download from, what to do with a hash mismatch, and how the cloud partners sidestep the whole problem.
Models
Upscaling AI images: ESRGAN models, hires-fix, tiled diffusion and when each one is right
The three families of upscaling and what they are for: fast pixel upscalers (4x-UltraSharp, Real-ESRGAN, SwinIR, DAT) for clean enlargement, hires-fix for adding detail at generation time, and tiled diffusion (Ultimate SD Upscale, SUPIR, Flux-based) for turning a 1024 render into a detailed 4K print; with denoise values, tile sizes and the models worth downloading.
Errors & fixes
GPU and VRAM guide for AI image and video: what each model needs and what to buy (or not)
How much graphics memory each model family really needs at usable speed (SD 1.5, SDXL, Flux, FLUX.2, Qwen Image, Z-Image, Wan 2.2, LTX-2, H3), what fp8 and GGUF quantisation buy you, why system RAM and disk matter too, a tier list of cards from 8 to 32 GB, Mac and AMD notes, and the point at which renting is cheaper than buying.
Prompting
Steps, CFG, samplers, schedulers and seeds: the generation settings explained with safe defaults
What each setting in the generation panel changes, the values that work per model family (SD 1.5, SDXL, Flux, Qwen, Z-Image, Wan), which samplers are worth knowing, why distilled models break the usual rules, and how to use seeds to change one thing at a time.
Models
Checkpoint vs LoRA vs embedding vs ControlNet: which file does what
The six kinds of model files you will meet, what each one changes, how big it is, where it goes and when to use it: checkpoints and fine-tunes, LoRAs and their LyCORIS cousins, textual inversion embeddings, ControlNets and IP-Adapters, VAEs and upscalers.
Models
How generative AI image models work: diffusion, latents and text encoders explained simply
The mechanics behind Stable Diffusion, SDXL, Flux and the video models without the maths: what noise has to do with it, why models work in a compressed latent space, what the text encoder and VAE do, what steps and guidance really change, and why training data decides what a model can draw.