piclets / src /lib /components /MonsterGenerator /MonsterGenerator.svelte
Fraser's picture
better bck rm
edc4c90
raw
history blame
24.2 kB
<script lang="ts">
import type { MonsterGeneratorProps, MonsterWorkflowState, CaptionType, CaptionLength, MonsterStats } from '$lib/types';
import type { PicletInstance } from '$lib/db/schema';
import UploadStep from './UploadStep.svelte';
import WorkflowProgress from './WorkflowProgress.svelte';
import MonsterResult from './MonsterResult.svelte';
import { removeBackground } from '$lib/utils/professionalImageProcessing';
import { saveMonster } from '$lib/db/monsters';
import { extractPicletMetadata } from '$lib/services/picletMetadata';
import { savePicletInstance } from '$lib/db/piclets';
interface Props extends MonsterGeneratorProps {}
let { joyCaptionClient, rwkvClient, fluxClient }: Props = $props();
let state: MonsterWorkflowState = $state({
currentStep: 'upload',
userImage: null,
imageCaption: null,
monsterConcept: null,
monsterStats: null,
imagePrompt: null,
monsterImage: null,
error: null,
isProcessing: false
});
// Custom prompt for joy-caption-alpha-two to generate everything in one step
const MONSTER_GENERATION_PROMPT = `Based on this image create a Pokémon-style monster that transforms the object into an imaginative creature. The monster should clearly be inspired by the object's appearance but reimagined as a living monster.
Guidelines:
- Take the object's key visual elements (colors, shapes, materials) incorporating all of them into a single creature design
- Add eyes (can be glowing, mechanical, multiple, etc.) positioned where they make sense
- Include limbs (legs, arms, wings, tentacles) that grow from or replace parts of the object
- Add a mouth, beak, or feeding apparatus if appropriate
- Add creature elements like tail, fins, claws, or horns where fitting
Include:
- A creative name that hints at the original object
- Physical description showing how the object becomes a creature
- Special abilities derived from the object's function
- Personality traits based on the object's purpose
Format your response as:
\`\`\`
# {monster name}
## Monster Visual Description
...
## Monster Lore
...
\`\`\``;
const IMAGE_GENERATION_PROMPT = (concept: string) => `Extract ONLY the visual appearance from this monster concept and describe it in one concise sentence:
"${concept}"
Focus on: colors, body shape, eyes, limbs, mouth, and key visual features. Omit backstory, abilities, and non-visual details.`;
async function importPiclet(picletData: PicletInstance) {
state.isProcessing = true;
state.currentStep = 'complete';
try {
// Save the imported piclet
const savedId = await savePicletInstance(picletData);
// Create a success state similar to generation
state.monsterImage = {
imageUrl: picletData.imageUrl,
imageData: picletData.imageData
};
// Show import success
state.isProcessing = false;
alert(`Successfully imported ${picletData.nickname || picletData.typeId}!`);
// Reset to allow another import/generation
setTimeout(() => reset(), 2000);
} catch (error) {
state.error = `Failed to import piclet: ${error}`;
state.isProcessing = false;
}
}
async function handleImageSelected(file: File) {
if (!joyCaptionClient || !rwkvClient || !fluxClient) {
state.error = "Services not connected. Please wait...";
return;
}
state.userImage = file;
state.error = null;
// Check if this is a piclet card with metadata
const picletData = await extractPicletMetadata(file);
if (picletData) {
// Import existing piclet
await importPiclet(picletData);
} else {
// Generate new piclet
startWorkflow();
}
}
async function startWorkflow() {
state.isProcessing = true;
try {
// Step 1: Generate monster concept with joy-caption (includes lore, visual description, and rarity)
await captionImage();
await new Promise(resolve => setTimeout(resolve, 100)); // Small delay for state update
// Step 2: Generate monster stats based on the concept
await generateStats();
await new Promise(resolve => setTimeout(resolve, 100)); // Small delay for state update
// Step 3: Generate monster image
await generateMonsterImage();
// Step 4: Auto-save the monster
await autoSaveMonster();
state.currentStep = 'complete';
} catch (err) {
console.error('Workflow error:', err);
// Check for GPU quota error
if (err && typeof err === 'object' && 'message' in err) {
const errorMessage = String(err.message);
if (errorMessage.includes('exceeded your GPU quota') || errorMessage.includes('GPU quota')) {
state.error = 'GPU quota exceeded! You need to sign in with Hugging Face for free GPU time, or upgrade to Hugging Face Pro for more quota.';
} else {
state.error = errorMessage;
}
} else if (err instanceof Error) {
state.error = err.message;
} else {
state.error = 'An unknown error occurred';
}
} finally {
state.isProcessing = false;
}
}
function handleAPIError(error: any): never {
console.error('API Error:', error);
// Check if it's a GPU quota error
if (error && typeof error === 'object' && 'message' in error) {
const errorMessage = String(error.message);
if (errorMessage.includes('exceeded your GPU quota') || errorMessage.includes('GPU quota')) {
throw new Error('GPU quota exceeded! You need to sign in with Hugging Face for free GPU time, or upgrade to Hugging Face Pro for more quota.');
}
throw new Error(errorMessage);
}
// Check if error has a different structure (like the status object from the logs)
if (error && typeof error === 'object' && 'type' in error && error.type === 'status') {
const statusError = error as any;
if (statusError.message && statusError.message.includes('GPU quota')) {
throw new Error('GPU quota exceeded! You need to sign in with Hugging Face for free GPU time, or upgrade to Hugging Face Pro for more quota.');
}
throw new Error(statusError.message || 'API request failed');
}
throw error;
}
async function captionImage() {
state.currentStep = 'captioning';
if (!joyCaptionClient || !state.userImage) {
throw new Error('Caption service not available or no image provided');
}
try {
const output = await joyCaptionClient.predict("/stream_chat", [
state.userImage, // input_image
"Descriptive", // caption_type
"very long", // caption_length
[], // extra_options
"", // name_input
MONSTER_GENERATION_PROMPT // custom_prompt
]);
const [prompt, caption] = output.data;
// The caption now contains the full monster concept with lore, visual description, and rarity
state.imageCaption = caption;
state.monsterConcept = caption; // Store as concept since it's the full monster details
console.log('Monster concept generated:', caption);
} catch (error) {
handleAPIError(error);
}
}
async function generateMonsterImage() {
state.currentStep = 'generating';
if (!fluxClient || !state.monsterConcept) {
throw new Error('Image generation service not available or no concept');
}
// Extract the visual description from the joy-caption output
const visualDescMatch = state.monsterConcept.match(/## Monster Visual Description\s*\n([\s\S]*?)(?=##|$)/);
if (visualDescMatch && visualDescMatch[1]) {
state.imagePrompt = visualDescMatch[1].trim();
console.log('Extracted visual description for image generation:', state.imagePrompt);
} else {
// Fallback: use zephyr to extract visual description
if (!rwkvClient) {
throw new Error('Text generation service not available for fallback');
}
const promptGenerationPrompt = IMAGE_GENERATION_PROMPT(state.monsterConcept);
const systemPrompt = "You are an expert at creating concise visual descriptions for image generation. Extract ONLY visual appearance details and describe them in ONE sentence (max 50 words). Focus on colors, shape, eyes, limbs, and distinctive features. Omit all non-visual information like abilities, personality, or backstory.";
console.log('Falling back to zephyr for visual description extraction');
try {
const output = await rwkvClient.predict("/chat", [
promptGenerationPrompt, // message
[], // chat_history
systemPrompt, // system_prompt
1024, // max_new_tokens
0.7, // temperature
0.95, // top_p
50, // top_k
1.0 // repetition_penalty
]);
state.imagePrompt = output.data[0];
} catch (error) {
handleAPIError(error);
}
}
if (!state.imagePrompt || state.imagePrompt.trim() === '') {
throw new Error('Failed to extract visual description');
}
try {
const output = await fluxClient.predict("/infer", [
`${state.imagePrompt}\nNow generate a Pokémon-Anime-style image of the monster in an idle pose with a white background. The monster should not be attacking or in motion. The full monster must be visible within the frame.`,
0, // seed
true, // randomizeSeed
1024, // width
1024, // height
4 // steps
]);
const [image, usedSeed] = output.data;
let url: string | undefined;
if (typeof image === "string") url = image;
else if (image && image.url) url = image.url;
else if (image && image.path) url = image.path;
if (url) {
// Process the image to remove background using professional AI method
console.log('Processing image for background removal...');
try {
const transparentBase64 = await removeBackground(url);
state.monsterImage = {
imageUrl: url,
imageData: transparentBase64,
seed: usedSeed,
prompt: state.imagePrompt
};
console.log('Background removal completed successfully');
} catch (processError) {
console.error('Failed to process image for background removal:', processError);
// Fallback to original image
state.monsterImage = {
imageUrl: url,
seed: usedSeed,
prompt: state.imagePrompt
};
}
} else {
throw new Error('Failed to generate monster image');
}
} catch (error) {
handleAPIError(error);
}
}
async function generateStats() {
state.currentStep = 'statsGenerating';
if (!rwkvClient || !state.monsterConcept) {
throw new Error('Text generation service not available or no concept');
}
// Default tier (will be set from the generated stats)
let tier: 'low' | 'medium' | 'high' | 'legendary' = 'medium';
// Extract monster name from the concept
const nameMatch = state.monsterConcept.match(/^#\s+(.+)$/m);
const monsterName = nameMatch ? nameMatch[1] : 'Unknown Monster';
// Extract only the Monster Lore section for description
const loreMatch = state.monsterConcept.match(/## Monster Lore\s*\n([\s\S]*?)(?=##|$)/);
const monsterLore = loreMatch ? loreMatch[1].trim() : '';
// Create stats prompt - only ask for battle-related stats
const statsPrompt = `Based on this monster concept, generate a JSON object with battle stats and abilities:
"${state.monsterConcept}"
First, assess the rarity of the original object based on real-world availability and value:
• COMMON: Everyday items everyone has (stationery, grass, rocks, basic furniture, common tools)
• UNCOMMON: Items that cost money but are widely available (electronics, appliances, vehicles, branded items)
• RARE: Expensive or specialized items (luxury goods, professional equipment, gold jewelry, antiques)
• LEGENDARY: Priceless or one-of-a-kind items (crown jewels, world wonders, famous artifacts, masterpiece art)
The output should be formatted as a JSON instance that conforms to the JSON schema below.
\`\`\`json
{
"properties": {
"rarity": {"type": "string", "enum": ["common", "uncommon", "rare", "legendary"], "description": "Rarity of the original object based on real-world availability and value"},
"height": {"type": "number", "minimum": 0.1, "maximum": 50.0, "description": "Height of the piclet in meters (e.g., 1.2, 0.5, 10.0)"},
"weight": {"type": "number", "minimum": 0.1, "maximum": 10000.0, "description": "Weight of the piclet in kilograms (e.g., 25.4, 150.0, 0.8)"},
"HP": {"type": "integer", "minimum": 0, "maximum": 100, "description": "Health/vitality stat (0=fragile, 100=incredibly tanky)"},
"defence": {"type": "integer", "minimum": 0, "maximum": 100, "description": "Defensive/armor stat (0=paper thin, 100=impenetrable fortress)"},
"attack": {"type": "integer", "minimum": 0, "maximum": 100, "description": "Physical attack power (0=harmless, 100=devastating force)"},
"speed": {"type": "integer", "minimum": 0, "maximum": 100, "description": "Movement and reaction speed (0=immobile, 100=lightning fast)"},
"specialPassiveTraitDescription": {"type": "string", "description": "Describe a passive ability that gives this monster a unique advantage in battle"},
"attackActionName": {"type": "string", "description": "Name of the monster's primary damage-dealing attack (e.g., 'Flame Burst', 'Toxic Bite')"},
"attackActionDescription": {"type": "string", "description": "Describe how this attack damages the opponent and any special effects"},
"buffActionName": {"type": "string", "description": "Name of the monster's self-enhancement ability (e.g., 'Iron Defense', 'Speed Boost')"},
"buffActionDescription": {"type": "string", "description": "Describe which stats are boosted and how this improves the monster's battle performance"},
"debuffActionName": {"type": "string", "description": "Name of the monster's enemy-weakening ability (e.g., 'Intimidate', 'Slow Poison')"},
"debuffActionDescription": {"type": "string", "description": "Describe which enemy stats are lowered and how this weakens the opponent"},
"specialActionName": {"type": "string", "description": "Name of the monster's ultimate move (one use per battle)"},
"specialActionDescription": {"type": "string", "description": "Describe this powerful finishing move and its dramatic effects in battle"}
},
"required": ["rarity", "height", "weight", "HP", "defence", "attack", "speed", "specialPassiveTraitDescription", "attackActionName", "attackActionDescription", "buffActionName", "buffActionDescription", "debuffActionName", "debuffActionDescription", "specialActionName", "specialActionDescription"]
}
\`\`\`
Base the HP, defence, attack, and speed stats on the rarity level:
- common: stats should be 10-40
- uncommon: stats should be 30-60
- rare: stats should be 50-80
- legendary: stats should be 70-100
Write your response within \`\`\`json\`\`\``;
const systemPrompt = "You are a game designer specializing in monster stats and abilities. You must ONLY output valid JSON that matches the provided schema exactly. Do not include any text before or after the JSON. Do not include null values in your JSON response. Your entire response should be wrapped in a ```json``` code block.";
console.log('Generating monster stats from concept');
try {
const output = await rwkvClient.predict("/chat", [
statsPrompt, // message
[], // chat_history
systemPrompt, // system_prompt
2048, // max_new_tokens
0.3, // temperature
0.95, // top_p
50, // top_k
1.0 // repetition_penalty
]);
console.log('Stats output:', output);
let jsonString = output.data[0];
// Extract JSON from the response (remove markdown if present)
let cleanJson = jsonString;
if (jsonString.includes('```')) {
const matches = jsonString.match(/```(?:json)?\s*([\s\S]*?)```/);
if (matches) {
cleanJson = matches[1];
} else {
// If no closing ```, just remove the opening ```json
cleanJson = jsonString.replace(/^```(?:json)?\s*/, '').replace(/```\s*$/, '');
}
}
try {
// Remove any trailing text after the JSON object
const jsonMatch = cleanJson.match(/^\s*\{[\s\S]*?\}\s*/);
if (jsonMatch) {
cleanJson = jsonMatch[0];
}
const parsedStats = JSON.parse(cleanJson.trim());
// Remove any extra fields not in our schema
const allowedFields = ['rarity', 'height', 'weight', 'HP', 'defence', 'attack', 'speed',
'specialPassiveTraitDescription', 'attackActionName', 'attackActionDescription',
'buffActionName', 'buffActionDescription', 'debuffActionName', 'debuffActionDescription',
'specialActionName', 'specialActionDescription', 'boostActionName', 'boostActionDescription',
'disparageActionName', 'disparageActionDescription'];
for (const key in parsedStats) {
if (!allowedFields.includes(key)) {
delete parsedStats[key];
}
}
// Map rarity to tier
if (parsedStats.rarity) {
const tierMap: { [key: string]: 'low' | 'medium' | 'high' | 'legendary' } = {
'common': 'low',
'uncommon': 'medium',
'rare': 'high',
'legendary': 'legendary'
};
tier = tierMap[parsedStats.rarity.toLowerCase()] || 'medium';
}
// Add the name, description, and tier that we extracted/mapped
parsedStats.name = monsterName;
parsedStats.description = monsterLore;
parsedStats.tier = tier;
// Ensure numeric fields are actually numbers
const numericFields = ['HP', 'defence', 'attack', 'speed'];
for (const field of numericFields) {
if (parsedStats[field] !== undefined) {
// Convert string numbers to actual numbers
parsedStats[field] = parseInt(parsedStats[field]);
// Clamp to 0-100 range
parsedStats[field] = Math.max(0, Math.min(100, parsedStats[field]));
}
}
// Map field names from schema to interface
if (parsedStats.specialPassiveTraitDescription) {
parsedStats.specialPassiveTrait = parsedStats.specialPassiveTraitDescription;
delete parsedStats.specialPassiveTraitDescription;
}
// Handle potential old field names from LLM
if (parsedStats.boostActionName) {
parsedStats.buffActionName = parsedStats.boostActionName;
delete parsedStats.boostActionName;
}
if (parsedStats.boostActionDescription) {
parsedStats.buffActionDescription = parsedStats.boostActionDescription;
delete parsedStats.boostActionDescription;
}
if (parsedStats.disparageActionName) {
parsedStats.debuffActionName = parsedStats.disparageActionName;
delete parsedStats.disparageActionName;
}
if (parsedStats.disparageActionDescription) {
parsedStats.debuffActionDescription = parsedStats.disparageActionDescription;
delete parsedStats.disparageActionDescription;
}
const stats: MonsterStats = parsedStats;
state.monsterStats = stats;
console.log('Monster stats generated:', stats);
console.log('Monster stats JSON:', JSON.stringify(stats, null, 2));
} catch (parseError) {
console.error('Failed to parse JSON:', parseError, 'Raw output:', cleanJson);
throw new Error('Failed to parse monster stats JSON');
}
} catch (error) {
handleAPIError(error);
}
}
async function autoSaveMonster() {
if (!state.monsterImage || !state.imageCaption || !state.monsterConcept || !state.imagePrompt || !state.monsterStats) {
console.error('Cannot auto-save: missing required data');
return;
}
try {
// Create a clean copy of stats to ensure it's serializable
const cleanStats = JSON.parse(JSON.stringify(state.monsterStats));
const monsterData = {
name: state.monsterStats.name,
imageUrl: state.monsterImage.imageUrl,
imageData: state.monsterImage.imageData,
imageCaption: state.imageCaption,
concept: state.monsterConcept,
imagePrompt: state.imagePrompt,
stats: cleanStats
};
// Check for any non-serializable data
console.log('Checking monster data for serializability:');
console.log('- name type:', typeof monsterData.name);
console.log('- imageUrl type:', typeof monsterData.imageUrl);
console.log('- imageData type:', typeof monsterData.imageData, monsterData.imageData ? `length: ${monsterData.imageData.length}` : 'null/undefined');
console.log('- imageCaption type:', typeof monsterData.imageCaption);
console.log('- concept type:', typeof monsterData.concept);
console.log('- imagePrompt type:', typeof monsterData.imagePrompt);
console.log('- stats:', cleanStats);
const id = await saveMonster(monsterData);
console.log('Monster auto-saved with ID:', id);
} catch (err) {
console.error('Failed to auto-save monster:', err);
console.error('Monster data that failed to save:', {
name: state.monsterStats?.name,
hasImageUrl: !!state.monsterImage?.imageUrl,
hasImageData: !!state.monsterImage?.imageData,
hasStats: !!state.monsterStats
});
// Don't throw - we don't want to interrupt the workflow
}
}
function reset() {
state = {
currentStep: 'upload',
userImage: null,
imageCaption: null,
monsterConcept: null,
monsterStats: null,
imagePrompt: null,
monsterImage: null,
error: null,
isProcessing: false
};
}
</script>
<div class="monster-generator">
{#if state.currentStep !== 'upload'}
<WorkflowProgress currentStep={state.currentStep} error={state.error} />
{/if}
{#if state.currentStep === 'upload'}
<UploadStep
onImageSelected={handleImageSelected}
isProcessing={state.isProcessing}
/>
{:else if state.currentStep === 'complete'}
<MonsterResult workflowState={state} onReset={reset} />
{:else}
<div class="processing-container">
<div class="spinner"></div>
<p class="processing-text">
{#if state.currentStep === 'captioning'}
Creating monster concept from your image...
{:else if state.currentStep === 'statsGenerating'}
Generating battle stats...
{:else if state.currentStep === 'generating'}
Generating your monster...
{/if}
</p>
</div>
{/if}
</div>
<style>
.monster-generator {
width: 100%;
max-width: 1200px;
margin: 0 auto;
padding: 2rem;
}
.processing-container {
display: flex;
flex-direction: column;
align-items: center;
padding: 3rem 1rem;
}
.spinner {
width: 60px;
height: 60px;
border: 3px solid #f3f3f3;
border-top: 3px solid #007bff;
border-radius: 50%;
animation: spin 1s linear infinite;
margin-bottom: 2rem;
}
@keyframes spin {
0% { transform: rotate(0deg); }
100% { transform: rotate(360deg); }
}
.processing-text {
font-size: 1.2rem;
color: #333;
margin-bottom: 2rem;
}
</style>