ARGENAI

Self-taught

ARGENAI

Dec 2025 - Present • 8 mos

My 2-Year Self-Taught Prompting Journey (March 2024 – March 2026) Started March 2024 experimenting with text-only models. Quickly shifted focus when image generation became widely accessible: learned to write ultra-detailed, cinematic prompts with style references, lighting, composition, lens info, mood boards in text, negative prompts, aspect ratios, weights, and seed control. Mid-2024: mastered iterative refinement — generate → critique → upscale → inpaint → vary region → blend multiple outputs. Built reusable image prompt templates (cinematic, product, character sheet, surreal, retro-futurist, etc.). Late 2024–early 2025: video generation arrived → dove into motion prompts: camera moves (dolly zoom, orbiting, slow pan), timing cues, frame transitions, style consistency across shots, keyframe descriptions, motion intensity sliders, sound-synced pacing. Experimented with image-to-video, text-to-video, extend, loop, lip-sync. 2025: became obsessed with multi-modal workflows — generate base image → refine with inpainting → turn into short video clip → add motion → upscale → post-process. Created personal libraries: 400+ image styles, 250+ video motion presets, common fix patterns (artifact removal, coherence hacks). Now March 2026: ~60% of my daily prompting is visual. Maintain separate vaults for stills (1,200+ entries) and motion (700+). Daily habit = generate 10–20 images/videos + analyze what worked. Always test newest image/video frontier models the day they drop — each leap usually doubles coherence, prompt adherence, or motion quality with zero extra effort. Biggest lesson: the skill ceiling for visuals keeps rising fastest. Switching to the latest generator remains the single strongest move — often turns “pretty good” into “indistinguishable from pro VFX” without rewriting the prompt.