Enterprise briefs rarely fit one skill. They mix product stills, vertical UGC, a 16:9 hero film, and maybe an explainer beat, each with different generation models and credit math. Claude Opus 4.8 is the catalog default in the Picturesque Director LLM picker because it holds multi-step tool chains together while building defensible plan cards.
Claude Opus 4.8 sits in the higher credit usage tier. It is not the model for "give me five hooks." It is the model for ambiguous client constraints, connector-aware plans, and approval-ready campaign cards.
Picker overview: Director LLM guide. Comparison: Director LLM comparison. UGC tone: Claude Fable 5 UGC guide. Long paste: Kimi K3 brand bible guide.
What Opus 4.8 Does in Director
Opus orchestrates skills, tools, and generation routing in one thread:
- Marketing campaign skill for bundled deliverables
- UGC ad skill for 9:16 social cuts
- Cinematic ad skill for hero film beats
- Product shot skill for packshots on Nano Banana Pro
- Motion on Kling 3.0 or SeeDance 2.5
Credit Usage: Higher Tier Default
Claude Opus 4.8 shows Higher usage in the picker. GPT 5.4 Nano and Mini are Lower usage rows for draft phases.
| Campaign phase | LLM pick |
|---|
| Internal exploration | GPT 5.4 Mini or Nano |
|---|---|
| Multi-skill plan card | Claude Opus 4.8 |
| UGC hook polish only | Claude Fable 5 one turn |
| Brand bible paste | Kimi K3 then Opus for final |
*Default model verified August 2026 in `llmModels.ts`.*
Enterprise-Style Brief Template
Paste into Director with Opus 4.8 selected:
> Client: B2B SaaS analytics. Deliverables: (1) three 1:1 product stills, (2) two 9:16 UGC hooks under 20s, (3) one 16:9 cinematic hero under 30s. Credit ceiling: 800. Skills: product shot, UGC, cinematic. Prefer Nano Banana Pro stills, Kling 3.0 for social motion, SeeDance 2.5 for hero if under ceiling. Legal: no ROI guarantees in hooks. Plan only first.
Opus should return a step table with generation models, aspect ratios, and credit estimates before any render.
Plan Card Workflow
Step 1: Select Claude Opus 4.8 (default) in the LLM picker.
Step 2: Attach client brief, brand deck, or link list. State credit ceiling and deliverable matrix.
Step 3: Request plan-only response with per-step credits.
Step 4: Review tool routing. Confirm each skill path matches public skills landings.
Step 5: Approve generation. Watch separate pools: LLM credits spent so far vs Nano Banana Pro / Kling line items.
Step 6: Iterate failed steps in chat on Mini if the plan structure is fine; stay on Opus if tools misfire.
Opus vs Fable vs GPT for Campaigns
| Question | Opus 4.8 | Fable 5 | GPT 5.4 Mini |
|---|
| Multi-skill tool chains | Best fit | Good creative tone | Good drafts |
|---|---|---|---|
| Usage tier | Higher | Higher | Lower |
| Enterprise ambiguity | Strong | Expressive, less structural | Fast, may miss edge rules |
| Hook-only task | Overkill | Strong | Cheap |
Pairing Opus with Workflows
Director ships the client campaign once. Workflows encode the winning graph for repeat quarterly refreshes. Opus helps document node order from the approved plan:
> Export this approved plan as a bullet list suitable for a workflow doc: node order, model paths, aspect ratios.
Ops clones the graph without re-running Opus on every SKU refresh.
Multi-Skill Campaign Example
Enterprise SaaS launch on Opus 4.8:
| Deliverable | Skill | Generation target |
|---|
| Packshots ×3 | Product shot | Nano Banana Pro 1:1 |
|---|---|---|
| UGC hooks ×2 | UGC ad | Kling 3.0 9:16 |
| Hero film | Cinematic ad | SeeDance 2.5 16:9 |
| Explainer beat | Explainer video | Route per plan card |
When Opus Beats Fable and GPT
| Brief shape | Best LLM |
|---|
| One vertical UGC script | Fable 5 (UGC guide) |
|---|---|
| Twenty hook labels | Grok 4.5 (hooks guide) |
| 40-page paste onboarding | Kimi K3 then Opus (Kimi guide) |
| Bundled deliverable matrix | Opus 4.8 |
| Internal shot-list only | GPT 5.4 Nano (Nano guide) |
Client Approval Checklist
Before you approve an Opus plan card:
- [ ] Every step lists generation model and aspect ratio
- [ ] Credit subtotal under client ceiling
- [ ] Legal lines appear on correct beats, not in hooks
- [ ] Skill paths match public /skills/ landings
- [ ] LLM credits spent so far noted in thread
Opus Default in the Picker
Claude Opus 4.8 is the catalog default in `llmModels.ts`. New Director sessions start on Opus unless you change the picker. That is correct for complex campaigns but expensive for hook-only threads.
| First message type | Change default before sending? |
|---|
| Enterprise multi-skill brief | Keep Opus 4.8 |
|---|---|
| Hook brainstorm | Switch to Grok 4.5 or Nano first |
| Brand bible paste | Switch to Kimi K3 first |
Read next: Director LLM guide · Kimi brand bible · Nano drafts · Marketing campaign skill.
Post-Approval Iteration on Opus
After the client signs the Opus plan card:
- Small text tweaks: switch to GPT 5.4 Mini in the same thread
- Failed tool step on one skill: stay on Opus 4.8 for that fix
- New deliverable added mid-campaign: new Opus plan pass, do not bolt on via Nano
Limitations to Be Aware Of
Overkill for simple hooks. Use Grok 4.5 or Nano first (see Grok hooks guide).
Higher usage tier. Do not run Opus for micro-edits after the plan is approved.
Feature flag. Director tab required.
Beta skills. Skill slash commands and routing can change week to week.
No external Claude subscription. Opus bills through Picturesque credits.
Separate generation credits. Multi-skill plans can stack Kling, SeeDance, and still costs quickly. Respect the ceiling in the brief.
Try Claude Opus 4.8 for campaign plans on Picturesque
Build multi-skill plan cards, then approve Kling 3.0 and Nano Banana Pro steps.