piclets / src /lib /components /MonsterGenerator /MonsterGenerator.svelte
Fraser's picture
pmt
d7e0329
raw
history blame
22.2 kB
<script lang="ts">
import type { MonsterGeneratorProps, MonsterWorkflowState, CaptionType, CaptionLength, MonsterStats } from '$lib/types';
import UploadStep from './UploadStep.svelte';
import WorkflowProgress from './WorkflowProgress.svelte';
import MonsterResult from './MonsterResult.svelte';
import { makeWhiteTransparent } from '$lib/utils/imageProcessing';
import { saveMonster } from '$lib/db/monsters';
interface Props extends MonsterGeneratorProps {}
let { joyCaptionClient, rwkvClient, fluxClient }: Props = $props();
let state: MonsterWorkflowState = $state({
currentStep: 'upload',
userImage: null,
imageCaption: null,
monsterConcept: null,
imagePrompt: null,
monsterImage: null,
error: null,
isProcessing: false
});
// Prompt templates
const MONSTER_CONCEPT_PROMPT = (caption: string) => `Based on this image caption: "${caption}"
Create a Pokémon-style monster that IS the object in the image brought to life. The monster should physically resemble and be based on the actual object's appearance, not just its conceptual properties.
The monster must look like the object itself transformed into a creature.
Assess rarity: Common everyday objects = weak monsters (low stats). Unique/rare objects = powerful monsters (high stats).
Include:
- How the object's physical form translates into the monster's body
- How its original function becomes abilities
- Personality based on the object's purpose
- Name that references what the object is`;
const IMAGE_GENERATION_PROMPT = (concept: string) => `Convert this monster concept into a clear and succinct description of its appearance:
"${concept}"
Include all of its visual details, format the description as a single long sentence.`;
const MONSTER_STATS_PROMPT = (concept: string) => `Convert the following monster concept into a JSON object with stats:
"${concept}"
The output should be formatted as a JSON instance that conforms to the JSON schema below.
\`\`\`json
{
"properties": {
"name": {"type": "string", "description": "The monster's unique name"},
"description": {"type": "string", "description": "A brief physical description of the monster's appearance"},
"rarity": {"type": "integer", "minimum": 0, "maximum": 100, "description": "How rare/unique the monster is (0=very common, 100=legendary)"},
"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": ["name", "description", "rarity", "HP", "defence", "attack", "speed", "specialPassiveTraitDescription", "attackActionName", "attackActionDescription", "buffActionName", "buffActionDescription", "debuffActionName", "debuffActionDescription", "specialActionName", "specialActionDescription"]
}
\`\`\`
Remember to base the stats on how unique/powerful the original object was. Common objects should have lower stats, unique objects should have higher stats.
Write your response within \`\`\`json\`\`\``;
async function handleImageSelected(file: File) {
if (!joyCaptionClient || !rwkvClient || !fluxClient) {
state.error = "Services not connected. Please wait...";
return;
}
state.userImage = file;
state.error = null;
startWorkflow();
}
async function startWorkflow() {
state.isProcessing = true;
try {
// Step 1: Caption the image
await captionImage();
await new Promise(resolve => setTimeout(resolve, 100)); // Small delay for state update
// Step 2: Generate monster concept
await generateMonsterConcept();
await new Promise(resolve => setTimeout(resolve, 100)); // Small delay for state update
// Step 3: Generate monster stats
await generateStats();
await new Promise(resolve => setTimeout(resolve, 100)); // Small delay for state update
// Step 4: Generate image prompt
await generateImagePrompt();
await new Promise(resolve => setTimeout(resolve, 100)); // Small delay for state update
// Step 5: Generate monster image
await generateMonsterImage();
// Step 6: 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,
"Descriptive" as CaptionType,
"very long" as CaptionLength,
[], // extra_options
"", // name_input
"" // custom_prompt
]);
const [prompt, caption] = output.data;
state.imageCaption = caption;
console.log('Caption generated:', caption);
} catch (error) {
handleAPIError(error);
}
}
async function generateMonsterConcept() {
state.currentStep = 'conceptualizing';
if (!rwkvClient || !state.imageCaption) {
throw new Error('Text generation service not available or no caption');
}
const conceptPrompt = MONSTER_CONCEPT_PROMPT(state.imageCaption);
const systemPrompt = "You are a Pokémon-style monster designer. When creating monsters, they must physically resemble the actual object - the monster IS the object come to life, not just inspired by it. A car becomes a living car monster, a book becomes a living book monster, etc. The object's physical form should be clearly recognizable in the monster design.";
console.log('Generating monster concept with prompt:', conceptPrompt);
try {
const output = await rwkvClient.predict("/chat", [
conceptPrompt, // message
[], // chat_history
systemPrompt, // system_prompt
2048, // max_new_tokens
0.7, // temperature
0.95, // top_p
50, // top_k
1.0 // repetition_penalty
]);
console.log('Zephyr output:', output.data[0]);
let conceptText = output.data[0];
state.monsterConcept = conceptText;
console.log('Monster concept generated:', state.monsterConcept);
if (!state.monsterConcept || state.monsterConcept.trim() === '') {
throw new Error('Failed to generate monster concept - received empty response');
}
} catch (error) {
handleAPIError(error);
}
}
async function generateImagePrompt() {
state.currentStep = 'promptCrafting';
if (!rwkvClient || !state.monsterConcept) {
throw new Error('Text generation service not available or no concept');
}
const promptGenerationPrompt = IMAGE_GENERATION_PROMPT(state.monsterConcept);
const systemPrompt = "You are an expert at creating visual descriptions for image generation. Provide clear, detailed visual descriptions. Do not refer to 'the image' instead just describe all the relevant visual details. Ensure your response is a single descriptive sentance, do not respond with a long paragraph.";
console.log('Generating image prompt from concept');
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
]);
console.log('Image prompt output:', output);
let promptText = output.data[0];
state.imagePrompt = promptText;
console.log('Image prompt generated:', state.imagePrompt);
if (!state.imagePrompt || state.imagePrompt.trim() === '') {
throw new Error('Failed to generate image prompt - received empty response');
}
} catch (error) {
handleAPIError(error);
}
}
async function generateMonsterImage() {
state.currentStep = 'generating';
if (!fluxClient || !state.imagePrompt) {
throw new Error('Image generation service not available or no prompt');
}
try {
const output = await fluxClient.predict("/infer", [
`${state.imagePrompt}\nNow generate an 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 make white background transparent
console.log('Processing image for transparency...');
try {
const transparentBase64 = await makeWhiteTransparent(url);
state.monsterImage = {
imageUrl: url,
imageData: transparentBase64,
seed: usedSeed,
prompt: state.imagePrompt
};
} catch (processError) {
console.error('Failed to process image for transparency:', 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');
}
const statsPrompt = MONSTER_STATS_PROMPT(state.monsterConcept);
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. 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 - lower for structured output
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 = ['name', 'description', 'rarity', '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];
}
}
// Ensure numeric fields are actually numbers
const numericFields = ['rarity', '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'}
Analyzing your image...
{:else if state.currentStep === 'conceptualizing'}
Creating monster concept...
{:else if state.currentStep === 'statsGenerating'}
Generating battle stats...
{:else if state.currentStep === 'promptCrafting'}
Crafting generation prompt...
{: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>