Tips and tricks
Posted: Wed Mar 18, 2026 8:32 pm
Let’s call this things I wish I had known or thought about earlier.
Hybrid creation
Start with generating 3 versions of the next part of your story with [Prompt].
Feed that back to the LLM with something like:
We are going to be comparing three drafts of a section of a chapter. I am going to give you a prompt as background to judge the 3, but do not generate anything using this prompt: “[Insert Prompt]”. Tell when ready for 3 drafts.
Grok response: I'm ready for the three drafts now. Please provide Draft 1, Draft 2, and Draft 3 (in any order you prefer), and I'll compare them based on the provided canonical prompt, source materials, guardrails, style rules, continuity requirements, tone/voice fidelity, structure ratios, sensory variety, humor/satire freshness, progression accuracy, and overall execution of Section 2.
I gave it the 3 versions cut and paste with v1: in front of the first, v2: in front of the second, and v3: in front of the third.
Grok’s response started with “ Here is a detailed comparison of the three drafts (V1, V2, V3) for Chapter 8: Safe Word, Section 2, evaluated strictly against the canonical prompt's requirements: exact continuity pickup from Section 1's ending text, adjusted arc/progression beats, mandatory style ratios (≥40% internal monologue/reflection with ≥15% fresh humorous/cynical/self-deprecating “whore-moans” satire; ≥20% deepening trust/craving for Julian’s framework; ≥30% varied sensory descriptions with 16–18+ entirely new phrasings—no reuse from prior chapters/Section 1), ~30% external action + dialogue (Julian’s quieter/rougher/protective-firm speech), first-person present tense from Elena’s POV, no breeding terminology in Elena’s thoughts/descriptions (only possible possessive implications in Julian’s dialogue), no removal of collar (wrists only unclipped), fresh satire on major beats (edging cycles, clamps, crop, repositioning, finger penetration, growls), no 3+ paragraphs without Elena’s reaction/POV, and ending precisely after extended aftercare in lap with hazy possessive/trust bond and subtle foreshadowing of future high-leg/Protocol elements.” and went on from there.
My next direction is why I took you through the rabbit hole: “ Give me the most memorable, hottest (best sensory prose), funniest/best landing humor by hybridizing the three. Comply with structure ratios but note that Elena’s POV percentage and the sensory percentages were set as minimums. Stay compliant to the underlying prompt.”. And that is how I landed on the final version of 8.2.
To double check Grok was happy with the hybrid, I asked it to check its own work: “ Evaluate your hybrid against the 3 originals in summary table format.”
I won’t post the table, but it was not a surprise this instance of grok which drafted the hybrid thought the hybrid won.
Bottom line - the answers produced by one prompt can vary in quality and compliance with your rules. Running the same prompt multiple times and hybridizing the results can result in a stronger draft.
Hybrid creation
Start with generating 3 versions of the next part of your story with [Prompt].
Feed that back to the LLM with something like:
We are going to be comparing three drafts of a section of a chapter. I am going to give you a prompt as background to judge the 3, but do not generate anything using this prompt: “[Insert Prompt]”. Tell when ready for 3 drafts.
Grok response: I'm ready for the three drafts now. Please provide Draft 1, Draft 2, and Draft 3 (in any order you prefer), and I'll compare them based on the provided canonical prompt, source materials, guardrails, style rules, continuity requirements, tone/voice fidelity, structure ratios, sensory variety, humor/satire freshness, progression accuracy, and overall execution of Section 2.
I gave it the 3 versions cut and paste with v1: in front of the first, v2: in front of the second, and v3: in front of the third.
Grok’s response started with “ Here is a detailed comparison of the three drafts (V1, V2, V3) for Chapter 8: Safe Word, Section 2, evaluated strictly against the canonical prompt's requirements: exact continuity pickup from Section 1's ending text, adjusted arc/progression beats, mandatory style ratios (≥40% internal monologue/reflection with ≥15% fresh humorous/cynical/self-deprecating “whore-moans” satire; ≥20% deepening trust/craving for Julian’s framework; ≥30% varied sensory descriptions with 16–18+ entirely new phrasings—no reuse from prior chapters/Section 1), ~30% external action + dialogue (Julian’s quieter/rougher/protective-firm speech), first-person present tense from Elena’s POV, no breeding terminology in Elena’s thoughts/descriptions (only possible possessive implications in Julian’s dialogue), no removal of collar (wrists only unclipped), fresh satire on major beats (edging cycles, clamps, crop, repositioning, finger penetration, growls), no 3+ paragraphs without Elena’s reaction/POV, and ending precisely after extended aftercare in lap with hazy possessive/trust bond and subtle foreshadowing of future high-leg/Protocol elements.” and went on from there.
My next direction is why I took you through the rabbit hole: “ Give me the most memorable, hottest (best sensory prose), funniest/best landing humor by hybridizing the three. Comply with structure ratios but note that Elena’s POV percentage and the sensory percentages were set as minimums. Stay compliant to the underlying prompt.”. And that is how I landed on the final version of 8.2.
To double check Grok was happy with the hybrid, I asked it to check its own work: “ Evaluate your hybrid against the 3 originals in summary table format.”
I won’t post the table, but it was not a surprise this instance of grok which drafted the hybrid thought the hybrid won.
Bottom line - the answers produced by one prompt can vary in quality and compliance with your rules. Running the same prompt multiple times and hybridizing the results can result in a stronger draft.